From df55e73970edd3640d837f1a63e5002f12cca448 Mon Sep 17 00:00:00 2001
From: roy6732856 <roy6732856@gmail.com>
Date: Tue, 25 Aug 2026 22:04:21 +0800
Subject: [PATCH 1/5] feat(google): add quota-aware multi-account pool and
 failover routing

- Support quota threshold auto-switching for google-antigravity OAuth accounts
- Implement session affinity, cooldown tracking, and bounded failover routing
- Add CLI `ocx account auto-switch google-antigravity` controls
- Expose management API endpoints for Google account pool configuration
- Update multi-lingual reference docs for CLI, configuration, and management API
- Add comprehensive unit, integration, and request failover test suites
---
 .../fr/reference/cli/providers-accounts.md    |  88 ++-
 .../fr/reference/configuration/providers.md   |  35 +-
 .../docs/fr/reference/management-api.md       |  27 +-
 .../ja/reference/cli/providers-accounts.md    |  84 ++-
 .../ja/reference/configuration/providers.md   |  34 +-
 .../docs/ja/reference/management-api.md       |  28 +-
 .../ko/reference/cli/providers-accounts.md    |  83 ++-
 .../ko/reference/configuration/providers.md   |  33 +-
 .../docs/ko/reference/management-api.md       |  26 +-
 .../docs/reference/cli/providers-accounts.md  |  84 ++-
 .../docs/reference/configuration/providers.md |  33 +-
 .../content/docs/reference/management-api.md  |  26 +-
 .../ru/reference/cli/providers-accounts.md    |  85 ++-
 .../ru/reference/configuration/providers.md   |  34 +-
 .../docs/ru/reference/management-api.md       |  27 +-
 .../tr/reference/cli/providers-accounts.md    |  89 ++-
 .../tr/reference/configuration/providers.md   |  33 +-
 .../docs/tr/reference/management-api.md       |  28 +-
 .../zh-cn/reference/cli/providers-accounts.md |  81 ++-
 .../reference/configuration/providers.md      |  29 +-
 .../docs/zh-cn/reference/management-api.md    |  25 +-
 .../zh-tw/reference/cli/providers-accounts.md |  78 ++-
 .../reference/configuration/providers.md      |  29 +-
 .../docs/zh-tw/reference/management-api.md    |  25 +-
 src/cli/account-extended.ts                   |  70 ++-
 src/cli/registry.ts                           |   7 +-
 src/codex/pool-rotation.ts                    |  25 +-
 src/config.ts                                 |  24 +
 src/lib/state-store-registrations.ts          |   9 +
 src/oauth/google-antigravity-routing.ts       | 516 +++++++++++++++++
 src/providers/quota.ts                        |  97 ++--
 src/routing/analytics.ts                      |   1 +
 src/server/management/oauth-account-routes.ts |  81 ++-
 src/server/responses/core.ts                  | 268 ++++++++-
 src/types/config.ts                           |  10 +
 src/usage/log.ts                              |   2 +
 tests/account-pool-management-api.test.ts     | 247 ++++++++
 tests/cli-account.test.ts                     | 138 ++++-
 tests/config.test.ts                          |  74 +--
 ...e-antigravity-account-pool-request.test.ts | 537 ++++++++++++++++++
 tests/google-antigravity-account-pool.test.ts | 287 ++++++++++
 tests/provider-account-quota.test.ts          |  61 ++
 42 files changed, 3321 insertions(+), 277 deletions(-)
 create mode 100644 src/oauth/google-antigravity-routing.ts
 create mode 100644 tests/google-antigravity-account-pool-request.test.ts
 create mode 100644 tests/google-antigravity-account-pool.test.ts

diff --git a/docs-site/src/content/docs/fr/reference/cli/providers-accounts.md b/docs-site/src/content/docs/fr/reference/cli/providers-accounts.md
index 4b9a4b710f..f993c59074 100644
--- a/docs-site/src/content/docs/fr/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/fr/reference/cli/providers-accounts.md
@@ -100,20 +100,26 @@ Répertoriez et changez de compte de fournisseur et de pools de clés API via le
 la surface est :
 
 ```text
-Usage: ocx account <list|current|use|refresh|auto-switch|priority|login|reauth|code|cancel|remove|add-key|reset-credits> ...
-
-list [provider]     Codex account pool, OAuth accounts and API keys (identifiers shown masked as the API returns them).
-current <provider>  Show the active account or key.
-use <provider> <id> Switch the active credential; 'main' selects the Codex App login.
-refresh <provider>  Force-refresh Codex or provider quota reports.
-auto-switch <provider> <on|off|status|threshold N>  Control the Codex pool threshold.
-priority <provider> <id|main> [first|earlier|normal|later|last|-100..100|reset]  Selection order; omit the value to read it.
-remove <provider> <id> --yes  Remove a stored account or key after an existence check.
-add-key <provider> [--label <label>]  Add a key read only from piped stdin.
-login/reauth/code/cancel  Run browser or manual-code auth from a headless shell.
-reset-credits <id|main> [--consume --yes]  Inspect or consume Codex reset credits.
-Switching the active account takes effect immediately; running threads move on their next request, and in-flight requests keep the account they captured.
-A selection-order change applies from the next unbound request and never moves a bound thread.
+Usage:
+  ocx account list [provider] [--json] [--all]
+  ocx account current <provider> [--json]
+  ocx account use <provider> <account-or-key-id|main> [--json]
+  ocx account refresh <provider> [--json]
+  ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]
+  ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]
+  ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]
+  ocx account remove <provider> <account-or-key-id|main> --yes [--json]
+  ocx account clear-cooldown <provider> <account-id|main> [--json]
+  ocx account add-key <provider> [--label <label>] [--json]
+  ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]
+  ocx account login <provider> [--id <account-id>] [--reauth] [--code -] [--no-wait] [--json]
+  ocx account code <provider> [--flow <flow-id>] [--json]   (reads the code from stdin)
+  ocx account cancel <provider> [--flow <flow-id>] [--json]
+  ocx account reset-credits <account-id|main> [--consume --yes] [--json]
+  ocx account main <doctor|list|register|add|switch|recover> ...
+
+List and switch provider accounts and API-key pools (masked output only).
+'main' selects the Codex App login for the openai account pool.
 ```
 
 Toutes les sous-commandes nécessitent que le proxy soit en cours d'exécution ; le CLI résout automatiquement son port d'exécution enregistré.
@@ -203,14 +209,28 @@ renvoient 1 ; une sonde de quota en amont qui échoue ou expire produit plutôt
 
 ### `ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]`
 
-Contrôle uniquement le groupe de comptes Codex `openai`. `on` règle 80 %, `off` règle 0 %, `status` lit la
-valeur actuelle et `threshold <n>` accepte un entier de 0 à 100. Les autres fournisseurs et les valeurs
-invalides entraînent le code de sortie 1. `--json` renvoie :
+Contrôle le pool Codex `openai` et les pools OAuth pris en charge `anthropic` et `google-antigravity`.
+Pour `openai`, `on` fixe le seuil à 80 % et `off` à 0 %. Pour les pools OAuth, `on` et `off`
+activent ou désactivent le pool tout en conservant son seuil. `threshold <n>` enregistre un entier de
+0 à 100 et active le pool si la valeur est différente de zéro. `status` lit la stratégie actuelle.
+Les fournisseurs non pris en charge et les valeurs invalides entraînent le code de sortie 1.
+`--json` renvoie :
 
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
 
+### `ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]`
+
+Définit un alias d'affichage sur un compte Codex, un compte OAuth ou une clé API existants ; passez
+`-` pour l'effacer. Un alias contient au plus 80 caractères imprimables. Le login Codex App réservé
+`main` ne peut pas être renommé : `ocx account alias openai main ...` se termine avec le code 1.
+`--json` renvoie :
+
+```text
+{ ok: true, provider, id, alias: string | null }
+```
+
 ### `ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]`
 
 Lit ou définit l’ordre de sélection d’un compte du groupe Codex : **une priorité plus élevée est utilisée plus tôt**.
@@ -245,7 +265,7 @@ l'actualisation de son catalogue de modèles reste en attente, la sortie humaine
 `ocx sync` conseils de récupération sur stderr. `--json` garde la sortie standard analysable et transporte
 `catalogRefreshPending: true` dans l'état de connexion terminé sans avertissement humain.
 
-### `ocx account remove <provider> <id|main> --yes [--json]`
+### `ocx account remove <provider> <account-or-key-id|main> --yes [--json]`
 
 Cette suppression gardée et non interactive nécessite `--yes`. Avant de supprimer, il vérifie que l'identifiant
 existe; un identifiant manquant quitte 1 sans envoyer DELETE. La connexion principale Codex App ne peut pas être supprimée, donc
@@ -263,6 +283,18 @@ Les formes de réussite et d’échec sont :
 déjà enregistré; la sortie humaine imprime des conseils de récupération génériques `ocx sync` sur stderr et quitte toujours 0.
 Les enveloppes de retrait de compte OAuth et de clé API n'obtiennent pas ce champ.
 
+### `ocx account clear-cooldown <provider> <account-id|main> [--json]`
+
+Efface une temporisation d'exécution du pool Codex `openai` ou des pools OAuth pris en charge
+`anthropic` et `google-antigravity`. `main` n'est accepté que pour `openai` et devient `__main__` dans
+le JSON. Un fournisseur non valide ou un compte inconnu produit une réponse API 400 et la CLI se
+termine avec le code 1. Pour un compte réel sans temporisation active, l'opération est idempotente :
+l'API renvoie 200, la CLI se termine avec le code 0 et `cleared` vaut `false`. `--json` renvoie exactement :
+
+```text
+{ ok: true, provider, id, cleared: boolean }
+```
+
 ### `ocx account add-key <provider> [--label <label>] [--json]`
 
 Ajoute et active une clé pour un fournisseur de clés API. La clé est lue uniquement à partir de non-TTY piped/redirected
@@ -276,6 +308,26 @@ security find-generic-password -w openrouter | ocx account add-key openrouter --
 
 `--json` renvoie `{ ok: true, id: string | null, label?: string }` et n'inclut jamais la clé.
 
+### `ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]`
+
+Importe une exportation de comptes de taille bornée. La version 1 accepte uniquement le fournisseur
+`google-antigravity` au format `cockpit-tools`, depuis exactement une source, `--file` ou `--stdin`.
+Le JSON en ligne et les arguments positionnels supplémentaires sont refusés avant l'examen de leur
+contenu. Gardez les exportations confidentielles, puis supprimez-les ou sécurisez-les après usage. La
+sortie JSON est le résumé sans secret ci-dessous. Toute entrée en échec ou non prise en charge fait se
+terminer la commande avec le code 1 ; sinon elle se termine avec le code 0.
+
+```text
+{
+  totalCount,
+  importedCount,
+  updatedCount,
+  failedCount,
+  unsupportedCount,
+  results: Array<{ index, status, code }>
+}
+```
+
 ### `ocx account reset-credits <id|main> [--consume --yes]`
 
 Inspectez Codex réinitialiser les crédits d'un compte. Consommer un crédit est destructeur et nécessite à la fois
diff --git a/docs-site/src/content/docs/fr/reference/configuration/providers.md b/docs-site/src/content/docs/fr/reference/configuration/providers.md
index 822c7ba47c..9bb593b002 100644
--- a/docs-site/src/content/docs/fr/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/fr/reference/configuration/providers.md
@@ -84,7 +84,6 @@ sauvegarde dont le contenu diffère, puis réécrit en identifiants sans préfix
 | `modelContextWindows?` | `Record<string, number>` | Valeurs de repli ou plafonds de contexte par modèle. Ils remplacent `contextWindow` : une fenêtre inconnue utilise la valeur configurée, tandis que des métadonnées actives plus faibles restent déterminantes. |
 | `modelInputModalities?` | `Record<string, string[]>` | Conseils de saisie par modèle tels que `["text"]` ou `["text", "image"]`. |
 | `modelMaxInputTokens?` | `Record<string, number>` | Limites d'entrée maximales positives par modèle utilisées pour les conseils de compactage automatique du catalogue. |
-| `modelAutoCompactTokenLimits?` | `Record<string, number>` | Budgets souples de compactage automatique par modèle, sous forme d'entiers sûrs positifs. Ils peuvent uniquement abaisser l'enveloppe effective de 90 % du contexte ou de l'entrée maximale et sont omis lorsqu'aucune fenêtre de contexte faisant autorité n'est connue. Pour le fournisseur canonique `openai`, les clés doivent être les identifiants exacts de modèles natifs pris en charge, sans préfixe de fournisseur ni de sélecteur de compte. PATCH fusionne les entrées ; `null` supprime une clé, tandis que `null` pour le champ entier efface la table. Ces marqueurs `null` sont réservés à PATCH. |
 | `defaultMaxOutputTokens?` | `number` | Solution de secours `openai-chat` à l’échelle du fournisseur lorsque le client omet `max_output_tokens`. |
 | `modelMaxOutputTokens?` | `Record<string, number>` | Budgets de repli `openai-chat` positifs par modèle ; les correspondances exactes ou par motif priment sur la valeur par défaut du fournisseur. |
 | `modelCosts?` | `Record<string, Cost4>` | Prix affichés par modèle (USD par 1M de jetons), indexés par l'identifiant exact du modèle en amont de ce fournisseur — et non par un identifiant de fournisseur ni par une étiquette routée `provider/model`, par exemple `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`. Tout identifiant de modèle constitue une clé valide : les fournisseurs personnalisés peuvent cibler n'importe quel point de terminaison compatible avec OpenAI au moyen de l'adaptateur `openai-chat`, et les identifiants de fournisseur locaux ou internes fonctionnent même s'ils sont absents des catalogues intégrés. Les prix configurés par l'utilisateur priment sur les catalogues intégrés dans les estimations des pages Journaux (`~$`) et Utilisation. Les entrées historiques sont recalculées à partir de la surcharge actuelle ; modifier un prix peut donc changer les totaux antérieurs. L'ordre de repli est le suivant : `modelCosts` défini par l'utilisateur → catalogue jawcode → surcharge des prix attendus → repli propre au fournisseur au niveau du modèle. Une entrée entièrement nulle passe à la source suivante. Chaque tarif doit être un nombre fini positif ou nul, inférieur ou égal à 1 000 000 (USD par 1M de jetons) ; les lignes hors plage sont rejetées par l'interface de gestion et ignorées au chargement. Ces valeurs servent uniquement à l'estimation lors de l'affichage : les surcharges n'affectent jamais le routage, la sélection des comptes, les quotas ni la facturation. |
@@ -207,6 +206,40 @@ Laissez cette option désactivée, sauf si vous comprenez les risques liés aux
 préférez le changement manuel avec `ocx account use anthropic <id>`.
 :::
 
+### `googleAntigravityAccountPool` (expérimental)
+
+Cette option regroupe des comptes OAuth Google Antigravity uniquement pour le fournisseur Cloud Code
+Assist `google-antigravity`. Elle est désactivée par défaut. Le pool ne fournit jamais d'identifiants à
+Google AI Studio, Vertex AI, une autre entrée de fournisseur ou une route utilisant une clé API.
+
+| Clé | Type | Par défaut | Description |
+| --- | --- | --- | --- |
+| `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | Active l'affinité de session locale au processus et un basculement borné après une réponse finale 429 ou une réponse 402 remontée. |
+| `googleAntigravityAccountPool.autoSwitchThreshold?` | `number` | `80` | Seuil d'épuisement de 0 à 100. `quota` et `fill-first` utilisent l'utilisation maximale en cache correspondant à la famille `Gem` ou `Cla` du modèle demandé ; une utilisation inconnue conserve le compte actif sain. `0` désactive le changement selon l'utilisation, pas la reprise après échec. |
+| `googleAntigravityAccountPool.strategy?` | `"quota" \| "round-robin" \| "fill-first"` | `"quota"` | Stratégie pour les sessions nouvelles ou sans liaison. |
+| `googleAntigravityAccountPool.stickyLimit?` | `number` | `1` | Nombre de liaisons de nouvelles sessions conservées sur une sélection `round-robin`. Plage 1–100 ; sans effet sur les autres stratégies. |
+
+Les stratégies conservent l'affinité d'une session tant que son compte reste admissible. Pour une
+session nouvelle ou sans liaison, `quota` conserve le compte actif sain lorsque l'utilisation pertinente
+est inconnue ou inférieure au seuil, puis choisit le compte admissible dont l'utilisation connue est la
+plus faible. `round-robin` répartit uniformément les liaisons et n'emploie pas le seuil pour la rotation
+ordinaire. `fill-first` remplit le compte actif jusqu'à sa temporisation, son indisponibilité ou le seuil
+d'épuisement, puis passe au compte admissible suivant.
+
+Chaque envoi résout un instantané OAuth lié à cette génération, avec le jeton bearer et le `projectId`
+Cloud Code Assist. Une réponse finale 429 ou une réponse 402 remontée met le compte en temporisation,
+efface son affinité et reconstruit la requête pour un autre instantané admissible. Sans `Retry-After`
+exploitable, la temporisation par défaut est de 60 secondes ; une valeur amont analysée est plafonnée à
+15 minutes. Une requête autorise au plus trois basculements, soit quatre envois amont au total. Si tous
+les comptes admissibles sont en temporisation, le client reçoit un 429 avec le premier `Retry-After`
+connu.
+
+:::caution[Résilience opérationnelle, pas contournement de quota]
+Utilisez ce pool pour récupérer après des défaillances transitoires, pas pour contourner les quotas ou
+les contrôles du fournisseur. L'automatisation multicomptes peut enfreindre les conditions du fournisseur ;
+laissez cette fonction désactivée si vous n'acceptez pas ce risque.
+:::
+
 ### Formes d'enregistrement gérées
 
 Les entrées `apiKeys[]` contiennent les chaînes `id`, `name`, la valeur `key` générée et la date ISO `createdAt`.
diff --git a/docs-site/src/content/docs/fr/reference/management-api.md b/docs-site/src/content/docs/fr/reference/management-api.md
index 46a0eb0278..61be7f9c81 100644
--- a/docs-site/src/content/docs/fr/reference/management-api.md
+++ b/docs-site/src/content/docs/fr/reference/management-api.md
@@ -174,14 +174,35 @@ d’abord et soumettez le résumé renvoyé. Préférez la quarantaine lorsqu’
 | `POST /api/oauth/logout` | Supprimer les informations d'identification du fournisseur sélectionné | 400 fournisseur inconnu ; `oauth_mutation_busy` |
 | `GET, DELETE /api/oauth/accounts` | Répertorier les comptes masqués ou supprimer un compte | 400 invalide provider/id ; 404 compte manquant ; `oauth_mutation_busy` |
 | `PUT /api/oauth/accounts/active` | Sélectionnez le compte OAuth actif | 400 invalide provider/account ; `oauth_mutation_busy` |
-| `GET, PUT, PATCH /api/oauth/accounts/pool` | Lire ou mettre à jour la stratégie du pool OAuth Anthropic | 400 fournisseur non Anthropic ou stratégie invalide |
-| `POST /api/oauth/accounts/clear-cooldown` | Effacer le temps de recharge d'un compte OAuth | 400 invalide provider/account |
+| `GET, PUT, PATCH /api/oauth/accounts/pool` | Lire ou mettre à jour la stratégie du pool OAuth Anthropic ou Google Antigravity | 400 fournisseur non pris en charge ou stratégie invalide |
+| `POST /api/oauth/accounts/clear-cooldown` | Effacer la temporisation d'un compte OAuth | 400 fournisseur non valide ou compte inconnu ; un compte connu sans temporisation renvoie 200 avec `{ ok: true, cleared: false }` |
 | `PUT /api/oauth/accounts/alias` | Définir ou supprimer un alias de compte OAuth | 400 invalide provider/account/alias |
 | `GET, POST, DELETE /api/providers/keys` | Répertorier les clés de fournisseur masquées, en ajouter ou en activer une, ou en supprimer une | 400 saisie invalide ; 404 fournisseur ou clé manquante |
 | `PUT /api/providers/keys/active` | Sélectionnez la clé active d'un fournisseur | 400 saisie invalide ; 404 provider/key manquant |
 | `PUT /api/providers/keys/alias` | Définir ou supprimer un alias de clé de fournisseur | 400 saisie invalide ; 404 provider/key manquant |
 | `GET, POST, PATCH, DELETE /api/keys` | Répertorier, créer, modifier ou supprimer les clés d'admission du plan de données | 400 corps ou identifiant invalide ; 404 clé manquante |
 
+#### Contrat des pools OAuth par fournisseur
+
+`GET /api/oauth/accounts/pool?provider=anthropic|google-antigravity` exige le paramètre de requête
+`provider`. Sa réponse contient directement les champs `provider`, `enabled`, `autoSwitchThreshold`
+(par défaut `80`), `strategy` (normalisé en `quota`, `round-robin` ou `fill-first`), `stickyLimit`
+(normalisé en entier de 1 à 100) et `experimental: true`.
+
+`PUT` et `PATCH /api/oauth/accounts/pool` exigent un objet JSON avec `provider` et acceptent les
+champs facultatifs `enabled` (booléen), `autoSwitchThreshold` (entier de 0 à 100), `strategy`
+(`quota`, `round-robin` ou `fill-first`) et `stickyLimit` (entier de 1 à 100). Les champs omis
+conservent leur valeur actuelle ou par défaut. En cas de succès, la réponse contient directement
+`{ ok: true, provider, enabled, autoSwitchThreshold, strategy, stickyLimit, experimental: true }`,
+sans enveloppe `config`. Un corps mal formé ou qui n'est pas un objet, un fournisseur non pris en
+charge ou un champ invalide renvoie une réponse d'erreur HTTP 400 générique.
+
+`POST /api/oauth/accounts/clear-cooldown` accepte `{ provider, accountId }`. Le fournisseur doit être
+`anthropic` ou `google-antigravity`, et le compte doit exister pour ce fournisseur. Un compte inconnu
+renvoie l'erreur HTTP 400 générique `account not found`. Un compte connu qui n'est pas en
+temporisation produit un succès idempotent : HTTP 200 avec `{ ok: true, cleared: false }`. Ces
+réponses ne contiennent jamais de jetons, d'adresses e-mail ni de clés complètes.
+
 Les réponses qui répertorient les identifiants sont délibérément masquées. Les jetons d'accès OAuth et les clés API complètes des
 fournisseurs ne sont pas renvoyés aux clients du tableau de bord.
 
@@ -245,7 +266,7 @@ Codex. Ses routes sont les suivantes :
 | `PUT /api/codex-auth/accounts/alias` | Définir ou supprimer un alias de compte | 400 invalide account/alias |
 | `PUT /api/codex-auth/accounts/pause` | Suspendre ou reprendre un compte | 400 invalide account/state ; 404 compte manquant |
 | `PUT /api/codex-auth/accounts/pause-exhausted` | Suspendre les comptes dont le quota est épuisé | Les échecs de verrouillage de mutation deviennent 503 |
-| `POST /api/codex-auth/accounts/clear-cooldown` | Effacer le temps de recharge d'exécution pour un compte ou tous les comptes | 400 identifiant invalide |
+| `POST /api/codex-auth/accounts/clear-cooldown` | Effacer la temporisation d'exécution d'un compte | 400 identifiant de compte non valide ou inconnu ; un compte connu sans temporisation renvoie 200 avec `{ ok: true, id, cleared: false }` |
 | `GET, PUT /api/codex-auth/active` | Lire ou sélectionner le compte actif | 400 compte invalide ou manquant ; 409 conflit avec un compte suspendu ou une ancienne ligne |
 | `PUT /api/codex-auth/auto-switch` | Définir le seuil de quota pour le changement automatique de compte | 400 seuil invalide |
 | `PUT, PATCH /api/codex-auth/pool-strategy` | Mettre à jour la stratégie de sélection du groupe de comptes Codex | 400 stratégie ou configuration invalide |
diff --git a/docs-site/src/content/docs/ja/reference/cli/providers-accounts.md b/docs-site/src/content/docs/ja/reference/cli/providers-accounts.md
index cab525623a..f50cafbd70 100644
--- a/docs-site/src/content/docs/ja/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/ja/reference/cli/providers-accounts.md
@@ -76,19 +76,26 @@ ocx login anthropic
 実行中のプロキシを介してプロバイダー アカウントと API キー プールを一覧表示し、切り替えます。出荷されたヘルプ画面は次のとおりです。
 
 ```text
-Usage: ocx account <list|current|use|refresh|auto-switch|priority|login|reauth|code|cancel|remove|add-key|reset-credits> ...
-
-list [provider]     Codex account pool, OAuth accounts and API keys (identifiers shown masked as the API returns them).
-current <provider>  Show the active account or key.
-use <provider> <id> Switch the active credential; 'main' selects the Codex App login.
-refresh <provider>  Force-refresh Codex or provider quota reports.
-auto-switch <provider> <on|off|status|threshold N>  Control the Codex pool threshold.
-priority <provider> <id|main> [first|earlier|normal|later|last|-100..100|reset]  Selection order; omit the value to read it.
-remove <provider> <id> --yes  Remove a stored account or key after an existence check.
-add-key <provider> [--label <label>]  Add a key read only from piped stdin.
-login/reauth/code/cancel  Run browser or manual-code auth from a headless shell.
-reset-credits <id|main> [--consume --yes]  Inspect or consume Codex reset credits.
-Codex pool selection applies to the next request after clearing existing affinity; in-flight requests keep their captured account.
+Usage:
+  ocx account list [provider] [--json] [--all]
+  ocx account current <provider> [--json]
+  ocx account use <provider> <account-or-key-id|main> [--json]
+  ocx account refresh <provider> [--json]
+  ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]
+  ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]
+  ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]
+  ocx account remove <provider> <account-or-key-id|main> --yes [--json]
+  ocx account clear-cooldown <provider> <account-id|main> [--json]
+  ocx account add-key <provider> [--label <label>] [--json]
+  ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]
+  ocx account login <provider> [--id <account-id>] [--reauth] [--code -] [--no-wait] [--json]
+  ocx account code <provider> [--flow <flow-id>] [--json]   (reads the code from stdin)
+  ocx account cancel <provider> [--flow <flow-id>] [--json]
+  ocx account reset-credits <account-id|main> [--consume --yes] [--json]
+  ocx account main <doctor|list|register|add|switch|recover> ...
+
+List and switch provider accounts and API-key pools (masked output only).
+'main' selects the Codex App login for the openai account pool.
 ```
 
 すべてのサブコマンドではプロキシが実行されている必要があります。 CLI は、記録されたランタイム ポートを自動解決します。操作が成功した場合は 0 で終了します。無効な使用法、不明なプロバイダーまたはアカウント/キー ID、到達不能なプロキシ、または API エラーが発生した場合は 1 で終了します。資格情報フィールドは、管理 API が返したとおりに表示されます (マスキングを含む)。生の API キーと OAuth トークンは決して返されません。表示の利便性は、ダッシュボードと同様にクライアント側で合成されます。`main` は、`openai` アカウント プール内の Codex アプリ ログインの CLI エイリアスであり、電子メールのない OAuth アカウントは `Account N` として表示され、プラン/ラベル列はプラン、マスクされた電子メール、ラベル、およびマスクされたキーにわたってフォールバックされます。
@@ -149,12 +156,27 @@ OAuth プロバイダーと API キー プロバイダーの場合、これに
 
 ### `ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]`
 
-`openai` Codex アカウント プールのみを制御します。 `on` は 80% を設定し、`off` は 0% を設定します。`status` は現在の値を読み取り、`threshold <n>` は 0 ～ 100 の整数を受け入れます。他のプロバイダーと無効な値は 1 を終了します。`--json` は次を返します。
+`openai` Codex アカウントプールと、対応する `anthropic` / `google-antigravity` OAuth プールを制御します。
+`openai` では `on` がしきい値を 80% に、`off` が 0% に設定します。OAuth プールでは、`on` / `off` は
+現在のしきい値を保持したままプールの有効・無効を切り替えます。`threshold <n>` は 0 から 100 までの
+整数を保存し、0 以外ならプールを有効にします。`status` は現在のポリシーを読み取ります。未対応の
+プロバイダーと無効な値は終了コード 1 になります。`--json` は次を返します。
 
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
 
+### `ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]`
+
+既存の Codex アカウント、OAuth アカウント、API キーに表示用エイリアスを設定します。解除するには
+`-` を渡します。エイリアスは印字可能な 80 文字以内です。予約済みの Codex App ログイン `main` は
+名前を変更できないため、`ocx account alias openai main ...` は終了コード 1 になります。`--json` は
+次を返します。
+
+```text
+{ ok: true, provider, id, alias: string | null }
+```
+
 ### `ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]`
 
 Codex pool のアカウント別選択順を読み書きします。**値が大きいほど先に使われ**、既定は `0`、範囲は
@@ -182,7 +204,7 @@ preemption が未バインドリクエストを直ちに引き上げます。既
 
 ヘッドレス シェルからブラウザベースまたは手動コードのアカウント認証を実行します。プロバイダー固有のコマンド形式には `ocx account --help` を使用します。Codex account login は保存済みでも catalog refresh が保留中なら成功終了し、human output の stderr に固定の `ocx sync` 案内を出します。`--json` は案内を混ぜず、完了 state に `catalogRefreshPending: true` を保持します。
 
-### `ocx account remove <provider> <id|main> --yes [--json]`
+### `ocx account remove <provider> <account-or-key-id|main> --yes [--json]`
 
 この保護された非対話型削除には `--yes` が必要です。削除する前に、ID が存在することが確認されます。 ID が欠落している場合は、DELETE を送信せずに 1 が終了します。メインの Codex App ログインは削除できないため、`remove openai main --yes` は拒否されます。削除後、ファミリーは再度読み取られます。固定された Codex アカウントを削除すると、ピンがクリアされ、自動選択に戻ります。 OAuth は最初に残ったアカウントを昇格させるか、何も報告しません。 API キー プールは、最初に残っているキーを昇格するか、何も報告しません。 `--json` の成功と失敗の形状は次のとおりです。
 
@@ -194,6 +216,18 @@ preemption が未バインドリクエストを直ちに引き上げます。既
 `catalogRefreshPending` は Codex 削除だけに含まれます。`true` でも削除は保存済みで、human output は
 stderr に `ocx sync` の案内を出して終了コード 0 のままです。OAuth account と API key の削除形状は変わりません。
 
+### `ocx account clear-cooldown <provider> <account-id|main> [--json]`
+
+`openai` Codex プール、または対応する `anthropic` / `google-antigravity` OAuth プールの
+実行時クールダウンを 1 件解除します。`main` は `openai` だけで使用でき、JSON では `__main__`
+に正規化されます。無効なプロバイダーまたは不明なアカウントは API で 400 となり、CLI は終了
+コード 1 になります。実在するアカウントにクールダウンがない場合は冪等な成功で、API は 200、
+CLI は終了コード 0、`cleared` は `false` です。`--json` は正確に次を返します。
+
+```text
+{ ok: true, provider, id, cleared: boolean }
+```
+
 ### `ocx account add-key <provider> [--label <label>] [--json]`
 
 API キー プロバイダーのキーを追加してアクティブ化します。キーは、非 TTY パイプ/リダイレクトされた標準入力からの読み取り専用です。インタラクティブ TTY 入力、空の入力、OAuth/Codex プロバイダー、および API エラー終了 1。キーがラベル内に表示される場合も含め、キーがエコーされることはありません。シークレット マネージャーまたはヒア文字列を使用することをお勧めします。
@@ -205,6 +239,26 @@ security find-generic-password -w openrouter | ocx account add-key openrouter --
 
 `--json` は `{ ok: true, id: string | null, label?: string }` を返しますが、キーは決して含まれません。
 
+### `ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]`
+
+サイズ制限付きのアカウント書き出しを取り込みます。バージョン 1 が受け付けるのは
+`google-antigravity` プロバイダーと `cockpit-tools` 形式だけで、入力元は `--file` または
+`--stdin` のどちらか一方を指定します。インライン JSON と余分な位置引数は内容を調べる前に
+拒否されます。書き出しは非公開に保ち、使用後は削除するか安全に保管してください。JSON 出力は
+次の秘密を含まない概要です。失敗または未対応のレコードが 1 件でもあれば終了コード 1、なければ
+終了コード 0 です。
+
+```text
+{
+  totalCount,
+  importedCount,
+  updatedCount,
+  failedCount,
+  unsupportedCount,
+  results: Array<{ index, status, code }>
+}
+```
+
 ### `ocx account reset-credits <id|main> [--consume --yes]`
 
 アカウントの Codex リセット クレジットを検査します。クレジットの消費は破壊的であり、`--consume` と `--yes` の両方が必要です。
diff --git a/docs-site/src/content/docs/ja/reference/configuration/providers.md b/docs-site/src/content/docs/ja/reference/configuration/providers.md
index f1210892a5..e7122dd2f2 100644
--- a/docs-site/src/content/docs/ja/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ja/reference/configuration/providers.md
@@ -72,7 +72,6 @@ account を削除しても mapping は保持され、同じ id を再追加す
 | `modelContextWindows?` | `Record<string, number>` | モデルごとのコンテキスト値および上限。`contextWindow` より優先され、ウィンドウが不明なら設定値を使い、より小さいライブメタデータがあればそちらが優先されます。 |
 | `modelInputModalities?` | `Record<string, string[]>` | `["text"]` や `["text", "image"]` などのモデルごとの入力ヒント。 |
 | `modelMaxInputTokens?` | `Record<string, number>` |カタログの自動圧縮ヒントに使用されるモデルごとの正の最大入力制限。 |
-| `modelAutoCompactTokenLimits?` | `Record<string, number>` | モデルごとの正の安全な整数によるソフト自動圧縮予算。実効値であるコンテキストまたは最大入力の 90% の上限を下げることだけができ、信頼できるコンテキストウィンドウが不明な場合は出力されません。canonical `openai` では、キーは provider や account-selector の接頭辞を含まない、サポート対象の正確なネイティブモデル ID でなければなりません。provider PATCH はエントリをマージし、キーを `null` にするとそのキーを削除し、フィールド全体を `null` にするとマップを消去します。これらの `null` tombstone は PATCH 専用です。 |
 | `defaultMaxOutputTokens?` | `number` |クライアントが `max_output_tokens` を省略した場合の、プロバイダー全体の `openai-chat` フォールバック。 |
 | `modelMaxOutputTokens?` | `Record<string, number>` |モデルごとの `openai-chat` フォールバック バジェットがプラスになります。正確な/パターン一致はプロバイダーのデフォルトを上回ります。 |
 | `modelCosts?` | `Record<string, Cost4>` | モデルごとの表示価格（100万トークンあたりの米ドル）。そのプロバイダーの正確なアップストリーム モデル ID をキーにします（プロバイダー識別子やルーティングされた `provider/model` ラベルではありません）。値は `input`, `output`, `cacheRead`, `cacheWrite` の 4 フィールドです（例: `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`）。組み込みカタログにないモデル ID も、任意の OpenAI 互換エンドポイントを対象とするカスタムプロバイダーや、ローカル・内部プロバイダーで有効です。ユーザー設定の価格は Logs の `~$` と Usage の見積もりで組み込みカタログより優先されます。過去のエントリも現在のオーバーレイで再計算されるため、価格を編集すると過去の合計が変わることがあります（フォールバック順: ユーザー設定 → jawcode カタログ → expected-price オーバーレイ → モデル別ベンダー価格）。全ゼロのエントリは次のソースにフォールバックします。各レートは 0 以上の有限数で、最大 1,000,000（100万トークンあたりの米ドル）です。範囲外の行は管理境界で拒否され、読み込み時に破棄されます。表示専用の見積もりであり、ルーティング・アカウント選択・クォータ・請求には影響しません。 |
@@ -169,6 +168,39 @@ affinity を維持します。これらの戦略は provider enforcement を回
 Anthropic アカウント ポリシーのリスクを理解していない限り、これは無効のままにしてください。不明な場合は、`ocx account use anthropic <id>` を手動で切り替えることをお勧めします。
 :::
 
+### `googleAntigravityAccountPool`（実験的）
+
+このオプトイン機能は、`google-antigravity` Cloud Code Assist プロバイダー専用に Google
+Antigravity OAuth アカウントをプールします。デフォルトでは無効です。このプールの認証情報が
+Google AI Studio、Vertex AI、別のプロバイダー項目、API キールートに使われることはありません。
+
+| キー | 型 | デフォルト | 説明 |
+| --- | --- | --- | --- |
+| `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | プロセスローカルなセッション親和性と、最終的な 429 または表面化した 402 に対する回数制限付きフェイルオーバーを有効にします。 |
+| `googleAntigravityAccountPool.autoSwitchThreshold?` | `number` | `80` | 0～100 の使用量消化しきい値です。`quota` と `fill-first` は、要求モデルの `Gem` または `Cla` ファミリーに関係するキャッシュ済み使用量の最大値を使います。使用量が不明なら正常なアクティブアカウントを維持します。`0` は使用量による切り替えだけを無効にし、障害回復は無効にしません。 |
+| `googleAntigravityAccountPool.strategy?` | `"quota" \| "round-robin" \| "fill-first"` | `"quota"` | 新規セッションまたは未バインドのセッションに使う戦略です。 |
+| `googleAntigravityAccountPool.stickyLimit?` | `number` | `1` | 1 回の `round-robin` 選択に維持する新規セッションのバインド数です。範囲は 1～100 で、ほかの戦略には影響しません。 |
+
+どの戦略でも、対象アカウントが適格である間は既存のセッション親和性を維持します。新規または
+未バインドのセッションでは、`quota` は関連使用量が不明またはしきい値未満なら正常なアクティブ
+アカウントを維持し、その後は既知使用量が最小の適格アカウントを選びます。`round-robin` は
+バインドを均等に分配し、通常のローテーションにはしきい値を使いません。`fill-first` はクールダウン、
+利用不能、消化しきい値のいずれかに達するまでアクティブアカウントを使い、その後で次の適格
+アカウントへ進みます。
+
+各送信では、bearer token と Cloud Code Assist の `projectId` を含む、世代に固定された OAuth
+スナップショットを 1 つ解決します。最終的な 429 または表面化した 402 を受けると、そのアカウントを
+クールダウンし、親和性を解除して、別の適格なスナップショットでリクエストを再構築します。有効な
+`Retry-After` がなければ既定のクールダウンは 60 秒で、解析できた上流値は 15 分を上限とします。
+1 リクエストにつきフェイルオーバーは最大 3 回、上流への送信は合計最大 4 回です。すべての適格
+アカウントがクールダウン中なら、クライアントには最も早い既知の `Retry-After` を付けて 429 を返します。
+
+:::caution[運用上の回復力であり、クォータ回避ではありません]
+このプールは一時的なアカウント障害からの回復に使い、クォータやプロバイダーの制限を回避するために
+使わないでください。複数アカウントの自動運用は利用規約に違反する可能性があります。そのリスクを
+受け入れない場合は無効のままにしてください。
+:::
+
 ### 管理されたレコードの形状
 
 `apiKeys[]` エントリには、`id`、`name`、生成された `key`、および ISO `createdAt` 文字列が含まれます。 `codexAccounts[]` エントリには `id`、`email`、および `isMain` が必要で、オプションの `plan`、`chatgptAccountId`、およびプライバシー セーフな `logLabel` が必要です。これらのレコードは通常、ダッシュボードで管理されます。
diff --git a/docs-site/src/content/docs/ja/reference/management-api.md b/docs-site/src/content/docs/ja/reference/management-api.md
index 7982a6341c..8a46882aa6 100644
--- a/docs-site/src/content/docs/ja/reference/management-api.md
+++ b/docs-site/src/content/docs/ja/reference/management-api.md
@@ -150,14 +150,34 @@ Authorization: Bearer <admin-token>
 | `POST /api/oauth/logout` |選択したプロバイダー資格情報を削除します | 400 不明なプロバイダー。 `oauth_mutation_busy` |
 | `GET, DELETE /api/oauth/accounts` |マスクされたアカウントを一覧表示するか、アカウントを 1 つ削除する | 400 無効なプロバイダー/ID。 404 アカウントがありません。 `oauth_mutation_busy` |
 | `PUT /api/oauth/accounts/active` |アクティブな OAuth アカウントを選択します | 400 無効なプロバイダー/アカウント。 `oauth_mutation_busy` |
-| `GET, PUT, PATCH /api/oauth/accounts/pool` | Anthropic OAuth プール ポリシーの読み取りまたは更新 | 400 非 Anthropic プロバイダーまたは無効なポリシー |
-| `POST /api/oauth/accounts/clear-cooldown` | 1 つの OAuth アカウントのランタイム クールダウンをクリアする | 400 無効なプロバイダー/アカウント |
+| `GET, PUT, PATCH /api/oauth/accounts/pool` | Anthropic または Google Antigravity OAuth プール ポリシーの読み取りまたは更新 | 400 未対応プロバイダーまたは無効なポリシー |
+| `POST /api/oauth/accounts/clear-cooldown` | 1 つの OAuth アカウントのランタイム クールダウンを解除する | 400 無効なプロバイダーまたは不明なアカウント。既知のアカウントにクールダウンがなければ 200 と `{ ok: true, cleared: false }` |
 | `PUT /api/oauth/accounts/alias` | OAuth アカウント エイリアスを設定またはクリアする | 400 無効なプロバイダー/アカウント/エイリアス |
-| `GET, POST, DELETE /api/providers/keys` |マスクされたプロバイダー キーを一覧表示し、1 つを追加/アクティブ化するか、1 つを削除します。 400 無効な入力。 404 プロバイダー/キーがありません |
+| `GET, POST, DELETE /api/providers/keys` |マスクされたプロバイダー キーを一覧表示し、1 つを追加/アクティブ化するか、1 つを削除します。 | 400 無効な入力。 404 プロバイダー/キーがありません |
 | `PUT /api/providers/keys/active` |プロバイダーのアクティブなキーを選択します | 400 無効な入力。 404 プロバイダー/キーがありません |
 | `PUT /api/providers/keys/alias` |プロバイダー キー エイリアスを設定またはクリアする | 400 無効な入力。 404 プロバイダー/キーがありません |
 | `GET, POST, PATCH, DELETE /api/keys` |データ プレーン アドミッション キーの一覧表示、作成、編集、または削除 | 400 無効な本文/ID。 404 キーがありません |
 
+#### プロバイダー単位の OAuth プール契約
+
+`GET /api/oauth/accounts/pool?provider=anthropic|google-antigravity` には `provider` クエリパラメーターが
+必要です。レスポンスのトップレベルには `provider`、`enabled`、`autoSwitchThreshold`（既定値 `80`）、
+`strategy`（`quota`、`round-robin`、`fill-first` のいずれかに正規化）、`stickyLimit`（1～100 の整数に
+正規化）、`experimental: true` が含まれます。
+
+`PUT` / `PATCH /api/oauth/accounts/pool` には `provider` を含む JSON オブジェクトが必要で、任意で
+`enabled`（boolean）、`autoSwitchThreshold`（0～100 の整数）、`strategy`（`quota`、`round-robin`、
+`fill-first`）、`stickyLimit`（1～100 の整数）を指定できます。省略したフィールドは現在値または既定値を
+保持します。成功時はトップレベルで `{ ok: true, provider, enabled, autoSwitchThreshold, strategy,
+stickyLimit, experimental: true }` を返し、`config` ラッパーはありません。不正形式またはオブジェクト
+以外の body、未対応の provider、不正なフィールドは、汎用の HTTP 400 エラーレスポンスになります。
+
+`POST /api/oauth/accounts/clear-cooldown` の body は `{ provider, accountId }` です。provider は
+`anthropic` または `google-antigravity` で、その account は同じ provider 内に存在する必要があります。
+不明な account は汎用の HTTP 400 `account not found` エラーになります。既知の account が cooldown
+中でなくても、HTTP 200 と `{ ok: true, cleared: false }` を返す冪等な成功です。これらのレスポンスには
+token、メールアドレス、完全な key は含まれません。
+
 資格情報リストの応答は意図的にマスクされます。 OAuth アクセス トークンと完全なプロバイダー API キーはダッシュボード クライアントに返されません。
 
 ### プロバイダー
@@ -210,7 +230,7 @@ Authorization: Bearer <admin-token>
 | `PUT /api/codex-auth/accounts/alias` |アカウント エイリアスの設定またはクリア | 400 無効なアカウント/エイリアス |
 | `PUT /api/codex-auth/accounts/pause` | 1 つのアカウントを一時停止または再開する | 400 無効なアカウント/状態。 404 アカウントが見つかりません |
 | `PUT /api/codex-auth/accounts/pause-exhausted` |クォータを使い果たしたアカウントを一時停止する |ミューテーションロックの失敗は 503 になります |
-| `POST /api/codex-auth/accounts/clear-cooldown` | 1 つのアカウントまたはすべてのアカウントのランタイム クールダウンをクリアする | 400 無効な ID |
+| `POST /api/codex-auth/accounts/clear-cooldown` | 1 つのアカウントのランタイム クールダウンを解除する | 400 無効または不明なアカウント ID。既知のアカウントにクールダウンがなければ 200 と `{ ok: true, id, cleared: false }` |
 | `GET, PUT /api/codex-auth/active` |アクティブなアカウントを読み取るか選択します | 400 アカウントが無効または欠落しています。 409 一時停止/レガシー行の競合 |
 | `PUT /api/codex-auth/auto-switch` |自動アカウント切り替えのクォータしきい値を設定する | 400 無効なしきい値 |
 | `PUT, PATCH /api/codex-auth/pool-strategy` | Codex アカウントプールの選択戦略を更新 | 400 無効な戦略/構成 |
diff --git a/docs-site/src/content/docs/ko/reference/cli/providers-accounts.md b/docs-site/src/content/docs/ko/reference/cli/providers-accounts.md
index 5b7bd66ca9..c2b8f5e7e5 100644
--- a/docs-site/src/content/docs/ko/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/ko/reference/cli/providers-accounts.md
@@ -76,19 +76,26 @@ ocx login anthropic
 실행 중인 프록시를 통해 제공자 계정과 API 키 풀을 나열하고 전환합니다. 제공되는 도움말 표면은 다음과 같습니다:
 
 ```text
-Usage: ocx account <list|current|use|refresh|auto-switch|priority|login|reauth|code|cancel|remove|add-key|reset-credits> ...
-
-list [provider]     Codex account pool, OAuth accounts and API keys (identifiers shown masked as the API returns them).
-current <provider>  Show the active account or key.
-use <provider> <id> Switch the active credential; 'main' selects the Codex App login.
-refresh <provider>  Force-refresh Codex or provider quota reports.
-auto-switch <provider> <on|off|status|threshold N>  Control the Codex pool threshold.
-priority <provider> <id|main> [first|earlier|normal|later|last|-100..100|reset]  Selection order; omit the value to read it.
-remove <provider> <id> --yes  Remove a stored account or key after an existence check.
-add-key <provider> [--label <label>]  Add a key read only from piped stdin.
-login/reauth/code/cancel  Run browser or manual-code auth from a headless shell.
-reset-credits <id|main> [--consume --yes]  Inspect or consume Codex reset credits.
-Codex pool selection applies to the next request after clearing existing affinity; in-flight requests keep their captured account.
+Usage:
+  ocx account list [provider] [--json] [--all]
+  ocx account current <provider> [--json]
+  ocx account use <provider> <account-or-key-id|main> [--json]
+  ocx account refresh <provider> [--json]
+  ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]
+  ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]
+  ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]
+  ocx account remove <provider> <account-or-key-id|main> --yes [--json]
+  ocx account clear-cooldown <provider> <account-id|main> [--json]
+  ocx account add-key <provider> [--label <label>] [--json]
+  ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]
+  ocx account login <provider> [--id <account-id>] [--reauth] [--code -] [--no-wait] [--json]
+  ocx account code <provider> [--flow <flow-id>] [--json]   (reads the code from stdin)
+  ocx account cancel <provider> [--flow <flow-id>] [--json]
+  ocx account reset-credits <account-id|main> [--consume --yes] [--json]
+  ocx account main <doctor|list|register|add|switch|recover> ...
+
+List and switch provider accounts and API-key pools (masked output only).
+'main' selects the Codex App login for the openai account pool.
 ```
 
 모든 하위 명령은 프록시가 실행 중이어야 합니다. CLI는 기록된 런타임 포트를 자동으로 찾습니다. 성공한 작업은 종료 코드 0으로 끝납니다. 잘못된 사용, 알 수 없는 제공자 또는 계정/키 id, 도달할 수 없는 프록시, API 실패는 종료 코드 1로 끝납니다. 자격 증명 필드는 관리 API가 반환한 그대로 표시됩니다(마스킹도 그대로 포함됩니다). 원시 API 키와 OAuth 토큰은 절대 반환하지 않습니다. 표시 편의 기능은 대시보드와 마찬가지로 클라이언트 쪽에서 합성합니다. `main`은 `openai` 계정 풀의 Codex App 로그인에 대한 CLI 별칭이고, 이메일이 없는 OAuth 계정은 `Account N`으로 표시되며, plan/label 열은 plan, 마스킹된 이메일, label, 마스킹된 키 순으로 대체합니다.
@@ -149,12 +156,27 @@ OAuth 및 API 키 제공자에는 제공자의 할당량 보고 엔드포인트
 
 ### `ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]`
 
-`openai` Codex 계정 풀만 제어합니다. `on`은 80%, `off`는 0%를 설정하고, `status`는 현재 값을 읽으며, `threshold <n>`은 0부터 100까지의 정수를 받습니다. 다른 제공자와 잘못된 값은 종료 코드 1로 끝납니다. `--json`은 다음을 반환합니다:
+`openai` Codex 계정 풀과 지원되는 `anthropic` / `google-antigravity` OAuth 풀을 제어합니다.
+`openai`에서 `on`은 임계값을 80%로, `off`는 0%로 설정합니다. OAuth 풀에서 `on` / `off`는
+기존 임계값을 유지하면서 풀의 활성화 여부를 전환합니다. `threshold <n>`은 0부터 100까지의
+정수를 저장하며, 0이 아니면 풀을 활성화합니다. `status`는 현재 정책을 읽습니다. 지원되지 않는
+제공자와 잘못된 값은 종료 코드 1로 끝납니다. `--json`은 다음을 반환합니다:
 
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
 
+### `ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]`
+
+기존 Codex 계정, OAuth 계정 또는 API 키에 표시 별칭을 설정합니다. 별칭을 지우려면 `-`를
+전달하세요. 별칭은 인쇄 가능한 문자 80자 이하여야 합니다. 예약된 Codex App 로그인 `main`은
+이름을 바꿀 수 없으므로 `ocx account alias openai main ...`은 종료 코드 1이 됩니다. `--json`은
+다음을 반환합니다.
+
+```text
+{ ok: true, provider, id, alias: string | null }
+```
+
 ### `ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]`
 
 Codex pool 계정 하나의 선택 순서를 읽거나 설정합니다. **값이 클수록 먼저** 쓰이고 기본값은 `0`,
@@ -181,7 +203,7 @@ Codex pool 계정 하나의 선택 순서를 읽거나 설정합니다. **값이
 
 헤드리스 셸에서 브라우저 기반 또는 수동 코드 계정 인증을 실행합니다. 제공자별 명령 형태는 `ocx account --help`를 보십시오. Codex account login이 저장되었지만 catalog refresh가 보류 중이면 성공으로 종료하고 human output의 stderr에 고정된 `ocx sync` 안내를 표시합니다. `--json`은 안내를 섞지 않고 완료 state의 `catalogRefreshPending: true`를 유지합니다.
 
-### `ocx account remove <provider> <id|main> --yes [--json]`
+### `ocx account remove <provider> <account-or-key-id|main> --yes [--json]`
 
 이 보호된 비대화형 삭제는 `--yes`를 요구합니다. 삭제하기 전에 id가 존재하는지 확인하며, 없는 id는 DELETE를 보내지 않고 종료 코드 1로 끝납니다. Codex App의 main 로그인은 제거할 수 없으므로 `remove openai main --yes`는 거부됩니다. 삭제 후에는 해당 계열을 다시 읽습니다. 고정된 Codex 계정을 제거하면 고정이 풀리고 자동 선택으로 돌아갑니다. OAuth는 남아 있는 첫 번째 계정으로 승격하거나 없다고 보고합니다. API 키 풀은 남아 있는 첫 번째 키로 승격하거나 없다고 보고합니다. `--json`의 성공 및 실패 형식은 다음과 같습니다:
 
@@ -193,6 +215,18 @@ Codex pool 계정 하나의 선택 순서를 읽거나 설정합니다. **값이
 `catalogRefreshPending`는 Codex 삭제에만 포함됩니다. `true`여도 삭제는 이미 저장되었으며 human output은
 stderr에 `ocx sync` 안내를 표시하고 종료 코드 0을 유지합니다. OAuth account와 API key 삭제 형식은 바뀌지 않습니다.
 
+### `ocx account clear-cooldown <provider> <account-id|main> [--json]`
+
+`openai` Codex 풀 또는 지원되는 `anthropic` / `google-antigravity` OAuth 풀에서 런타임 쿨다운
+하나를 지웁니다. `main`은 `openai`에서만 허용되며 JSON에서는 `__main__`으로 정규화됩니다.
+잘못된 공급자나 알 수 없는 계정은 API에서 400으로 거부되고 CLI는 종료 코드 1이 됩니다. 실제
+계정에 활성 쿨다운이 없으면 멱등 성공으로 처리되어 API는 200, CLI는 종료 코드 0을 반환하고
+`cleared`는 `false`입니다. `--json`은 정확히 다음을 반환합니다.
+
+```text
+{ ok: true, provider, id, cleared: boolean }
+```
+
 ### `ocx account add-key <provider> [--label <label>] [--json]`
 
 API 키 제공자에 키를 추가하고 활성화합니다. 키는 비TTY 파이프/리디렉션 stdin에서만 읽습니다. 대화형 TTY 입력, 빈 입력, OAuth/Codex 제공자, API 실패는 종료 코드 1로 끝납니다. 라벨 안에 들어 있더라도 키는 절대 출력되지 않습니다. 비밀 관리자나 here-string을 쓰는 편이 좋습니다:
@@ -204,6 +238,25 @@ security find-generic-password -w openrouter | ocx account add-key openrouter --
 
 `--json`은 `{ ok: true, id: string | null, label?: string }`를 반환하며 키를 절대 포함하지 않습니다.
 
+### `ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]`
+
+크기가 제한된 계정 내보내기를 가져옵니다. 버전 1은 `google-antigravity` 공급자와
+`cockpit-tools` 형식만 지원하며, `--file` 또는 `--stdin` 중 정확히 하나를 입력으로 받아들입니다.
+인라인 JSON과 추가 위치 인수는 내용을 검사하기 전에 거부됩니다. 내보내기 파일은 비공개로 유지하고
+사용 후 삭제하거나 안전하게 보관하세요. JSON 출력은 아래의 비밀 없는 요약입니다. 실패하거나
+지원되지 않는 레코드가 있으면 종료 코드 1, 그렇지 않으면 종료 코드 0입니다.
+
+```text
+{
+  totalCount,
+  importedCount,
+  updatedCount,
+  failedCount,
+  unsupportedCount,
+  results: Array<{ index, status, code }>
+}
+```
+
 ### `ocx account reset-credits <id|main> [--consume --yes]`
 
 계정의 Codex reset credits를 확인합니다. credit을 소비하는 동작은 파괴적이므로 `--consume`와 `--yes`를 둘 다 요구합니다.
diff --git a/docs-site/src/content/docs/ko/reference/configuration/providers.md b/docs-site/src/content/docs/ko/reference/configuration/providers.md
index ccacb0a94f..2edd267ea4 100644
--- a/docs-site/src/content/docs/ko/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ko/reference/configuration/providers.md
@@ -72,7 +72,6 @@ managed map을 활성화하면 privacy-safe selector를 만들고, 이후 계정
 | `modelContextWindows?` | `Record<string, number>` | 모델별 컨텍스트 값이자 상한입니다. `contextWindow`보다 우선하며, 창 크기를 알 수 없으면 설정값을 쓰고 더 작은 라이브 메타데이터가 있으면 그쪽을 따릅니다. |
 | `modelInputModalities?` | `Record<string, string[]>` | `["text"]` 또는 `["text", "image"]` 같은 모델별 입력 힌트입니다. |
 | `modelMaxInputTokens?` | `Record<string, number>` | 카탈로그 자동 압축 힌트에 쓰는 양수 모델별 최대 입력 한도입니다. |
-| `modelAutoCompactTokenLimits?` | `Record<string, number>` | 모델별 양의 안전 정수형 소프트 자동 압축 예산입니다. 유효한 컨텍스트 또는 최대 입력의 90% 한도를 낮출 수만 있으며, 신뢰할 수 있는 컨텍스트 창을 알 수 없으면 내보내지 않습니다. canonical `openai`에서는 키가 공급자나 계정 선택자 접두사가 없는 정확한 지원 네이티브 모델 ID여야 합니다. 공급자 PATCH는 항목을 병합하며, 키를 `null`로 지정하면 해당 키를 삭제하고 필드 전체를 `null`로 지정하면 맵을 지웁니다. 이 `null` tombstone은 PATCH에서만 사용할 수 있습니다. |
 | `defaultMaxOutputTokens?` | `number` | 클라이언트가 `max_output_tokens`를 생략했을 때 쓰는 공급자 전반의 `openai-chat` 폴백입니다. |
 | `modelMaxOutputTokens?` | `Record<string, number>` | 양수 모델별 `openai-chat` 폴백 예산입니다. 정확한 일치와 패턴 일치가 공급자 기본값보다 우선합니다. |
 | `modelCosts?` | `Record<string, Cost4>` | 모델별 표시 가격(100만 토큰당 USD). 해당 공급자의 정확한 업스트림 모델 ID를 키로 사용하며(공급자 식별자나 라우팅된 `provider/model` 레이블이 아님) 값은 `input`, `output`, `cacheRead`, `cacheWrite` 네 필드입니다(예: `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`). 커스텀 공급자는 `openai-chat` 어댑터로 임의의 OpenAI 호환 엔드포인트를 대상으로 할 수 있으며, 내장 카탈로그에 없는 로컬·내부 공급자 ID도 유효합니다. 사용자 구성 가격은 Logs `~$` 및 Usage 추정에서 내장 카탈로그보다 우선합니다. 기존 항목도 현재 오버레이로 다시 계산되므로 가격을 편집하면 과거 합계가 바뀔 수 있습니다(폴백 순서: 사용자 설정 → jawcode 카탈로그 → expected-price 오버레이 → 모델별 벤더 가격). 전부 0인 항목은 다음 소스로 폴백합니다. 각 요율은 0 이상의 유한한 숫자이며 최대 1,000,000(100만 토큰당 USD)입니다. 범위를 벗어난 행은 관리 경계에서 거부되고 로드 시 삭제됩니다. 표시 전용 추정이며 라우팅·계정 선택·할당량·청구에는 영향을 주지 않습니다. |
@@ -173,6 +172,38 @@ affinity 초기화 뒤의 기존 작업도 포함될 수 있습니다. 출력 
 Anthropic 계정 정책 위험을 이해하지 못한다면 이 기능은 꺼두십시오. 확신이 없으면 수동 `ocx account use anthropic <id>` 전환을 우선하십시오.
 :::
 
+### `googleAntigravityAccountPool`(실험적)
+
+이 선택 기능은 `google-antigravity` Cloud Code Assist 공급자에만 Google Antigravity OAuth
+계정을 풀로 묶습니다. 기본값은 비활성화입니다. 이 풀의 자격 증명은 Google AI Studio,
+Vertex AI, 다른 공급자 항목 또는 API 키 경로에 사용되지 않습니다.
+
+| 키 | 타입 | 기본값 | 설명 |
+| --- | --- | --- | --- |
+| `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | 프로세스 로컬 세션 결속과 최종 429 또는 노출된 402 응답에 대한 제한된 failover를 켭니다. |
+| `googleAntigravityAccountPool.autoSwitchThreshold?` | `number` | `80` | 0–100 범위의 사용량 소진 임계값입니다. `quota`와 `fill-first`는 요청 모델의 `Gem` 또는 `Cla` 계열과 관련된 캐시 사용량의 최댓값을 사용하며, 사용량이 알려지지 않으면 정상 활성 계정을 유지합니다. `0`은 사용량 기반 전환만 끄고 장애 복구는 끄지 않습니다. |
+| `googleAntigravityAccountPool.strategy?` | `"quota" \| "round-robin" \| "fill-first"` | `"quota"` | 새 세션 또는 아직 결속되지 않은 세션의 전략입니다. |
+| `googleAntigravityAccountPool.stickyLimit?` | `number` | `1` | 한 번의 `round-robin` 선택에 유지할 새 세션 결속 수입니다. 범위는 1–100이며 다른 전략에는 영향을 주지 않습니다. |
+
+모든 전략은 계정이 적격한 동안 기존 세션 결속을 유지합니다. 새 세션이나 미결속 세션에서 `quota`는
+관련 사용량이 알려지지 않았거나 임계값 미만이면 정상 활성 계정을 유지하고, 이후에는 알려진 사용량이
+가장 낮은 적격 계정을 선택합니다. `round-robin`은 결속을 균등하게 분배하며 일반 순환에는 임계값을
+사용하지 않습니다. `fill-first`는 쿨다운, 사용 불가 또는 소진 임계값에 이를 때까지 활성 계정을
+사용한 다음 다음 적격 계정으로 넘어갑니다.
+
+각 전송은 bearer token과 Cloud Code Assist `projectId`를 함께 담은 세대 고정 OAuth 스냅샷 하나를
+확정합니다. 최종 429 또는 노출된 402는 실패한 계정을 쿨다운시키고 결속을 지운 뒤 다른 적격
+스냅샷으로 요청을 다시 만듭니다. 사용할 수 있는 `Retry-After`가 없으면 기본 쿨다운은 60초이며,
+파싱된 업스트림 값은 최대 15분으로 제한됩니다. 요청 하나당 failover는 최대 3회, 업스트림 전송은
+총 4회까지 허용됩니다. 모든 적격 계정이 쿨다운 중이면 가장 이른 알려진 `Retry-After`와 함께 429를
+클라이언트에 반환합니다.
+
+:::caution[운영 복원력이며 할당량 우회가 아닙니다]
+이 풀은 일시적인 계정 장애를 복구하는 데 사용하고 할당량이나 공급자 제한을 회피하는 데 사용하지
+마십시오. 다중 계정 자동화는 공급자 약관을 위반할 수 있으므로 그 위험을 감수하지 않는다면 비활성화
+상태로 두십시오.
+:::
+
 ### 관리되는 레코드 구조
 
 `apiKeys[]` 항목에는 `id`, `name`, 생성된 `key`, ISO 형식 `createdAt` 문자열이 들어갑니다. `codexAccounts[]` 항목에는 `id`, `email`, `isMain`이 필요하고, 선택적으로 `plan`, `chatgptAccountId`, 개인정보를 해치지 않는 `logLabel`을 둘 수 있습니다. 이런 레코드는 보통 대시보드가 관리합니다.
diff --git a/docs-site/src/content/docs/ko/reference/management-api.md b/docs-site/src/content/docs/ko/reference/management-api.md
index b0ed3dac31..8aa5aa32a2 100644
--- a/docs-site/src/content/docs/ko/reference/management-api.md
+++ b/docs-site/src/content/docs/ko/reference/management-api.md
@@ -150,14 +150,34 @@ Authorization: Bearer <admin-token>
 | `POST /api/oauth/logout` | 선택된 provider 자격 증명을 제거합니다 | 400 알 수 없는 provider; `oauth_mutation_busy` |
 | `GET, DELETE /api/oauth/accounts` | 마스킹된 계정을 나열하거나 계정 하나를 제거합니다 | 400 잘못된 provider/id; 404 계정 없음; `oauth_mutation_busy` |
 | `PUT /api/oauth/accounts/active` | 활성 OAuth 계정을 선택합니다 | 400 잘못된 provider/account; `oauth_mutation_busy` |
-| `GET, PUT, PATCH /api/oauth/accounts/pool` | Anthropic OAuth pool policy를 읽거나 업데이트합니다 | 400 Anthropic이 아닌 provider 또는 잘못된 policy |
-| `POST /api/oauth/accounts/clear-cooldown` | OAuth 계정 하나의 런타임 cooldown을 지웁니다 | 400 잘못된 provider/account |
+| `GET, PUT, PATCH /api/oauth/accounts/pool` | Anthropic 또는 Google Antigravity OAuth pool policy를 읽거나 업데이트합니다 | 400 지원하지 않는 provider 또는 잘못된 policy |
+| `POST /api/oauth/accounts/clear-cooldown` | OAuth 계정 하나의 런타임 cooldown을 지웁니다 | 400 잘못된 provider 또는 알 수 없는 account; 알려진 account에 cooldown이 없으면 200과 `{ ok: true, cleared: false }` 반환 |
 | `PUT /api/oauth/accounts/alias` | OAuth 계정 alias를 설정하거나 지웁니다 | 400 잘못된 provider/account/alias |
 | `GET, POST, DELETE /api/providers/keys` | 마스킹된 provider key를 나열, 추가/활성화, 또는 제거합니다 | 400 잘못된 입력; 404 provider/key 없음 |
 | `PUT /api/providers/keys/active` | provider의 활성 key를 선택합니다 | 400 잘못된 입력; 404 provider/key 없음 |
 | `PUT /api/providers/keys/alias` | provider-key alias를 설정하거나 지웁니다 | 400 잘못된 입력; 404 provider/key 없음 |
 | `GET, POST, PATCH, DELETE /api/keys` | 데이터 평면 admission key를 나열, 생성, 수정, 또는 삭제합니다 | 400 잘못된 본문/id; 404 key 없음 |
 
+#### 제공자 범위 OAuth 풀 계약
+
+`GET /api/oauth/accounts/pool?provider=anthropic|google-antigravity`에는 `provider` 쿼리 매개변수가
+필수입니다. 응답의 최상위 필드는 `provider`, `enabled`, `autoSwitchThreshold`(기본값 `80`),
+`strategy`(`quota`, `round-robin`, `fill-first` 중 하나로 정규화), `stickyLimit`(1~100 정수로
+정규화), `experimental: true`입니다.
+
+`PUT` / `PATCH /api/oauth/accounts/pool`에는 `provider`가 포함된 JSON 객체가 필요하며, 선택적으로
+`enabled`(boolean), `autoSwitchThreshold`(0~100 정수), `strategy`(`quota`, `round-robin`,
+`fill-first`), `stickyLimit`(1~100 정수)을 받을 수 있습니다. 생략된 필드는 현재값 또는 기본값을
+유지합니다. 성공 응답은 최상위에 `{ ok: true, provider, enabled, autoSwitchThreshold, strategy,
+stickyLimit, experimental: true }`를 반환하며 `config` 래퍼는 없습니다. 잘못된 형식이나 객체가 아닌
+본문, 지원되지 않는 provider, 잘못된 필드는 일반 HTTP 400 오류 응답을 반환합니다.
+
+`POST /api/oauth/accounts/clear-cooldown`의 본문은 `{ provider, accountId }`입니다. provider는
+`anthropic` 또는 `google-antigravity`여야 하며 account는 해당 provider 안에 존재해야 합니다. 알 수
+없는 account는 일반 HTTP 400 `account not found` 오류를 반환합니다. 알려진 account가 cooldown 상태가
+아니어도 HTTP 200과 `{ ok: true, cleared: false }`를 반환하는 멱등 성공입니다. 이 응답에는 token,
+이메일 주소 또는 전체 key가 포함되지 않습니다.
+
 자격 증명 목록 응답은 의도적으로 마스킹됩니다. OAuth access token과 완전한 provider API key는 대시보드 클라이언트에 반환되지 않습니다.
 
 ### 제공자
@@ -210,7 +230,7 @@ Authorization: Bearer <admin-token>
 | `PUT /api/codex-auth/accounts/alias` | 계정 alias를 설정하거나 지웁니다 | 400 잘못된 account/alias |
 | `PUT /api/codex-auth/accounts/pause` | 계정 하나를 일시 중지하거나 재개합니다 | 400 잘못된 account/state; 404 누락된 account |
 | `PUT /api/codex-auth/accounts/pause-exhausted` | quota가 소진된 account를 일시 중지합니다 | mutation-lock 실패는 503이 됩니다 |
-| `POST /api/codex-auth/accounts/clear-cooldown` | account 하나 또는 모든 account의 runtime cooldown을 지웁니다 | 400 잘못된 id |
+| `POST /api/codex-auth/accounts/clear-cooldown` | account 하나의 runtime cooldown을 지웁니다 | 400 잘못되거나 알 수 없는 account id; 알려진 account에 cooldown이 없으면 200과 `{ ok: true, id, cleared: false }` 반환 |
 | `GET, PUT /api/codex-auth/active` | 활성 account를 읽거나 선택합니다 | 400 잘못되었거나 누락된 account; 409 paused/legacy-row 충돌 |
 | `PUT /api/codex-auth/auto-switch` | 자동 account 전환을 위한 quota threshold를 설정합니다 | 400 잘못된 threshold |
 | `PUT, PATCH /api/codex-auth/pool-strategy` | Codex account-pool 선택 전략을 업데이트합니다 | 400 잘못된 전략/구성 |
diff --git a/docs-site/src/content/docs/reference/cli/providers-accounts.md b/docs-site/src/content/docs/reference/cli/providers-accounts.md
index a5d48eaf33..4a0235d02d 100644
--- a/docs-site/src/content/docs/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/reference/cli/providers-accounts.md
@@ -98,20 +98,26 @@ List and switch provider accounts and API-key pools through the running proxy. T
 surface is:
 
 ```text
-Usage: ocx account <list|current|use|refresh|auto-switch|priority|login|reauth|code|cancel|remove|add-key|reset-credits> ...
-
-list [provider]     Codex account pool, OAuth accounts and API keys (identifiers shown masked as the API returns them).
-current <provider>  Show the active account or key.
-use <provider> <id> Switch the active credential; 'main' selects the Codex App login.
-refresh <provider>  Force-refresh Codex or provider quota reports.
-auto-switch <provider> <on|off|status|threshold N>  Control the Codex pool threshold.
-priority <provider> <id|main> [first|earlier|normal|later|last|-100..100|reset]  Selection order; omit the value to read it.
-remove <provider> <id> --yes  Remove a stored account or key after an existence check.
-add-key <provider> [--label <label>]  Add a key read only from piped stdin.
-login/reauth/code/cancel  Run browser or manual-code auth from a headless shell.
-reset-credits <id|main> [--consume --yes]  Inspect or consume Codex reset credits.
-Switching the active account takes effect immediately; running threads move on their next request, and in-flight requests keep the account they captured.
-A selection-order change applies from the next unbound request and never moves a bound thread.
+Usage:
+  ocx account list [provider] [--json] [--all]
+  ocx account current <provider> [--json]
+  ocx account use <provider> <account-or-key-id|main> [--json]
+  ocx account refresh <provider> [--json]
+  ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]
+  ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]
+  ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]
+  ocx account remove <provider> <account-or-key-id|main> --yes [--json]
+  ocx account clear-cooldown <provider> <account-id|main> [--json]
+  ocx account add-key <provider> [--label <label>] [--json]
+  ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]
+  ocx account login <provider> [--id <account-id>] [--reauth] [--code -] [--no-wait] [--json]
+  ocx account code <provider> [--flow <flow-id>] [--json]   (reads the code from stdin)
+  ocx account cancel <provider> [--flow <flow-id>] [--json]
+  ocx account reset-credits <account-id|main> [--consume --yes] [--json]
+  ocx account main <doctor|list|register|add|switch|recover> ...
+
+List and switch provider accounts and API-key pools (masked output only).
+'main' selects the Codex App login for the openai account pool.
 ```
 
 All subcommands require the proxy to be running; the CLI auto-resolves its recorded runtime port.
@@ -215,14 +221,25 @@ instead (exit 0), matching the dashboard's quota bars.
 
 ### `ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]`
 
-Controls only the `openai` Codex account pool. `on` sets 80%, `off` sets 0%, `status` reads the current
-value, and `threshold <n>` accepts an integer from 0 through 100. Other providers and invalid values
-exit 1. `--json` returns:
+Controls the `openai` Codex pool and the supported `anthropic` and `google-antigravity` OAuth pools.
+For Codex, `on` sets 80% and `off` sets 0%. For OAuth pools, `on`/`off` toggle the pool while retaining
+its threshold; `threshold <n>` stores the integer from 0 through 100 and enables the pool when nonzero.
+`status` reads the current policy. Other providers and invalid values exit 1. `--json` returns:
 
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
 
+### `ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]`
+
+Sets a display alias on an existing Codex account, OAuth account, or API key; pass `-` to clear it.
+Aliases are at most 80 printable characters. The reserved `main` Codex App login cannot be renamed, so
+`ocx account alias openai main ...` exits 1. `--json` returns:
+
+```text
+{ ok: true, provider, id, alias: string | null }
+```
+
 ### `ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]`
 
 Reads or sets one Codex pool account's selection order: **higher is used earlier**, the default is
@@ -259,7 +276,7 @@ its model-catalog refresh remains pending, human output still exits successfully
 `ocx sync` recovery guidance on stderr. `--json` keeps stdout parseable and carries
 `catalogRefreshPending: true` in the completed login state without the human warning.
 
-### `ocx account remove <provider> <id|main> --yes [--json]`
+### `ocx account remove <provider> <account-or-key-id|main> --yes [--json]`
 
 This guarded, non-interactive deletion requires `--yes`. Before deleting, it verifies that the id
 exists; a missing id exits 1 without sending DELETE. The main Codex App login cannot be removed, so
@@ -277,6 +294,18 @@ success and failure shapes are:
 already saved; human output prints generic `ocx sync` recovery guidance on stderr and still exits 0.
 OAuth-account and API-key removal envelopes do not gain this field.
 
+### `ocx account clear-cooldown <provider> <account-id|main> [--json]`
+
+Clears one runtime cooldown in the `openai` Codex pool or the supported `anthropic` and
+`google-antigravity` OAuth pools. `main` is accepted only for `openai` and is normalized to `__main__`
+in JSON. An invalid provider or unknown account is rejected by the API with 400 and the CLI exits 1.
+A real account with no active cooldown is an idempotent success: the API returns 200, the CLI exits 0,
+and `cleared` is `false`. `--json` returns exactly:
+
+```text
+{ ok: true, provider, id, cleared: boolean }
+```
+
 ### `ocx account add-key <provider> [--label <label>] [--json]`
 
 Adds and activates a key for an API-key provider. The key is read only from non-TTY piped/redirected
@@ -290,6 +319,25 @@ security find-generic-password -w openrouter | ocx account add-key openrouter --
 
 `--json` returns `{ ok: true, id: string | null, label?: string }` and never includes the key.
 
+### `ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]`
+
+Imports a bounded account export. Version 1 accepts only provider `google-antigravity` with format
+`cockpit-tools`, from exactly one of `--file` or `--stdin`; inline JSON and extra positional arguments
+are rejected before their contents are inspected. Keep exports private and remove or secure them after
+use. JSON output is the secret-free summary below. Any failed or unsupported record makes the command
+exit 1; otherwise it exits 0.
+
+```text
+{
+  totalCount,
+  importedCount,
+  updatedCount,
+  failedCount,
+  unsupportedCount,
+  results: Array<{ index, status, code }>
+}
+```
+
 ### `ocx account reset-credits <id|main> [--consume --yes]`
 
 Inspect Codex reset credits for an account. Consuming a credit is destructive and requires both
diff --git a/docs-site/src/content/docs/reference/configuration/providers.md b/docs-site/src/content/docs/reference/configuration/providers.md
index 782effab52..144e312652 100644
--- a/docs-site/src/content/docs/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/reference/configuration/providers.md
@@ -86,7 +86,6 @@ differing backup and rewrites known legacy namespaced selected ids to bare ids.
 | `modelContextWindows?` | `Record<string, number>` | Per-model context fallbacks/caps. These override `contextWindow`: an unknown window uses the configured value, while smaller live metadata remains authoritative. |
 | `modelInputModalities?` | `Record<string, string[]>` | Per-model input hints such as `["text"]` or `["text", "image"]`. |
 | `modelMaxInputTokens?` | `Record<string, number>` | Positive per-model max input limits used for catalog auto-compaction hints. |
-| `modelAutoCompactTokenLimits?` | `Record<string, number>` | Positive safe-integer per-model soft auto-compaction budgets. Values can only lower the effective 90%-of-context/max-input envelope and are omitted when no authoritative context window is known. For canonical `openai`, keys must be exact supported native model IDs without provider or account-selector prefixes. Provider PATCH merges entries; set a key to `null` to delete it or the whole field to `null` to clear the map. These `null` tombstones are PATCH-only. |
 | `defaultMaxOutputTokens?` | `number` | Provider-wide `openai-chat` fallback when the client omits `max_output_tokens`. |
 | `modelMaxOutputTokens?` | `Record<string, number>` | Positive per-model `openai-chat` fallback budgets; exact/pattern matches beat the provider default. |
 | `modelCosts?` | `Record<string, Cost4>` | Per-model display prices (USD per 1M tokens), keyed by that provider's exact upstream model id — not a provider identifier or a routed `provider/model` label, e.g. `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`. Any model id is a valid key — custom providers may target any OpenAI-compatible endpoint through the `openai-chat` adapter, and local or internal provider ids work even when they are absent from the built-in catalogs. User-configured prices win over the built-in catalogs in the Logs `~$` and Usage estimates; historical entries are repriced from the current overlay, so editing a price can move past totals. The fallback order is user `modelCosts` → exact official correction → jawcode catalog → expected-price overlay → model-level vendor fallback, and an all-zero entry falls through to the next source in that sequence. Each rate must be a non-negative finite number at most 1,000,000 (USD per 1M tokens); out-of-range rows are rejected by the management boundary and dropped on load. Display-time estimation only: overlays never affect routing, account selection, quotas, or billing. |
@@ -295,6 +294,38 @@ Leave this disabled unless you understand Anthropic account policy risk. Prefer
 `ocx account use anthropic <id>` switching when unsure.
 :::
 
+### `googleAntigravityAccountPool` (experimental)
+
+This opt-in pools Google Antigravity OAuth accounts for the `google-antigravity` Cloud Code Assist
+provider only. It is disabled by default. The pool never supplies credentials to Google AI Studio,
+Vertex AI, another provider entry, or an API-key route.
+
+| Key | Type | Default | Description |
+| --- | --- | --- | --- |
+| `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | Enable process-local session affinity and bounded failover on a final 429 or surfaced 402 response. |
+| `googleAntigravityAccountPool.autoSwitchThreshold?` | `number` | `80` | Usage drain threshold from 0–100. `quota` and `fill-first` use the maximum cached usage relevant to the request model's `Gem` or `Cla` family; unknown usage keeps a healthy active account. `0` disables usage-driven switching, not failure recovery. |
+| `googleAntigravityAccountPool.strategy?` | `"quota" \| "round-robin" \| "fill-first"` | `"quota"` | Strategy for new or otherwise unbound sessions. |
+| `googleAntigravityAccountPool.stickyLimit?` | `number` | `1` | New-session binds retained on one `round-robin` selection. Range 1–100; it does not affect the other strategies. |
+
+The strategies preserve an eligible session affinity once it exists. For a new or unbound session,
+`quota` keeps the healthy active account while its relevant usage is unknown or below the threshold,
+then selects the eligible account with the lowest known usage. `round-robin` distributes binds evenly
+and does not use the threshold for ordinary rotation. `fill-first` keeps filling the active account
+until cooldown, unavailability, or the drain threshold, then advances to the next eligible account.
+
+Each dispatch resolves one generation-bound OAuth snapshot containing both the bearer token and Cloud
+Code Assist `projectId`. A final 429 or surfaced 402 cools the failed account, clears its affinity, and
+rebuilds the request for another eligible snapshot. The default cooldown is 60 seconds when no usable
+`Retry-After` is present; a parsed upstream value is capped at 15 minutes. One request permits at most
+three failovers (four total upstream dispatches). If every eligible account is cooling, the client
+receives 429 with the earliest known `Retry-After`.
+
+:::caution[Operational resilience, not quota bypass]
+Use this pool to recover from transient account failures, not to evade quotas or provider enforcement.
+Multi-account automation can violate provider terms; keep the feature disabled unless you accept that
+risk.
+:::
+
 ### `oauthAccountFailover`
 
 Rotates to another logged-in account of the same provider when one is rate-limited, for OAuth
diff --git a/docs-site/src/content/docs/reference/management-api.md b/docs-site/src/content/docs/reference/management-api.md
index bef307c3d5..7fa6d9a1ff 100644
--- a/docs-site/src/content/docs/reference/management-api.md
+++ b/docs-site/src/content/docs/reference/management-api.md
@@ -174,14 +174,34 @@ first and submit the returned digest. Prefer quarantine when recovery may be nee
 | `POST /api/oauth/logout` | Remove the selected provider credential | 400 unknown provider; `oauth_mutation_busy` |
 | `GET, DELETE /api/oauth/accounts` | List masked accounts or remove one account | 400 invalid provider/id; 404 account missing; `oauth_mutation_busy` |
 | `PUT /api/oauth/accounts/active` | Select the active OAuth account | 400 invalid provider/account; `oauth_mutation_busy` |
-| `GET, PUT, PATCH /api/oauth/accounts/pool` | Read or update Anthropic OAuth pool policy | 400 non-Anthropic provider or invalid policy |
-| `POST /api/oauth/accounts/clear-cooldown` | Clear one OAuth account's runtime cooldown | 400 invalid provider/account |
+| `GET, PUT, PATCH /api/oauth/accounts/pool` | Read or update Anthropic or Google Antigravity OAuth pool policy | 400 unsupported provider or invalid policy |
+| `POST /api/oauth/accounts/clear-cooldown` | Clear one OAuth account's runtime cooldown | 400 invalid provider or unknown account; a known account with no cooldown returns 200 with `{ ok: true, cleared: false }` |
 | `PUT /api/oauth/accounts/alias` | Set or clear an OAuth account alias | 400 invalid provider/account/alias |
 | `GET, POST, DELETE /api/providers/keys` | List masked provider keys, add/activate one, or remove one | 400 invalid input; 404 provider/key missing |
 | `PUT /api/providers/keys/active` | Select a provider's active key | 400 invalid input; 404 provider/key missing |
 | `PUT /api/providers/keys/alias` | Set or clear a provider-key alias | 400 invalid input; 404 provider/key missing |
 | `GET, POST, PATCH, DELETE /api/keys` | List, create, edit, or delete data-plane admission keys | 400 invalid body/id; 404 key missing |
 
+#### Provider-scoped OAuth pool contract
+
+`GET /api/oauth/accounts/pool?provider=anthropic|google-antigravity` requires the `provider` query
+parameter. Its top-level response fields are `provider`, `enabled`, `autoSwitchThreshold` (default
+`80`), `strategy` (normalized to `quota`, `round-robin`, or `fill-first`), `stickyLimit` (normalized
+to the integer range 1–100), and `experimental: true`.
+
+`PUT` and `PATCH /api/oauth/accounts/pool` require a JSON object with `provider` and accept optional
+`enabled` (boolean), `autoSwitchThreshold` (integer 0–100), `strategy` (`quota`, `round-robin`, or
+`fill-first`), and `stickyLimit` (integer 1–100). Omitted fields retain their current or default
+values. Success returns top-level `{ ok: true, provider, enabled, autoSwitchThreshold, strategy,
+stickyLimit, experimental: true }`; there is no `config` wrapper. A malformed or non-object body,
+an unsupported provider, or any invalid field returns a generic HTTP 400 error response.
+
+`POST /api/oauth/accounts/clear-cooldown` takes `{ provider, accountId }`. The provider must be
+`anthropic` or `google-antigravity`, and the account must exist within that provider. An unknown
+account returns a generic HTTP 400 `account not found` error. A known account that is not cooled down
+is an idempotent success: HTTP 200 with `{ ok: true, cleared: false }`. These responses never include
+tokens, email addresses, or complete keys.
+
 Credential list responses are deliberately masked. OAuth access tokens and complete provider API
 keys are not returned to dashboard clients.
 
@@ -245,7 +265,7 @@ manager. Its routes are:
 | `PUT /api/codex-auth/accounts/alias` | Set or clear an account alias | 400 invalid account/alias |
 | `PUT /api/codex-auth/accounts/pause` | Pause or resume one account | 400 invalid account/state; 404 missing account |
 | `PUT /api/codex-auth/accounts/pause-exhausted` | Pause accounts whose quota is exhausted | Mutation-lock failures become 503 |
-| `POST /api/codex-auth/accounts/clear-cooldown` | Clear runtime cooldown for one account or all accounts | 400 invalid id |
+| `POST /api/codex-auth/accounts/clear-cooldown` | Clear one account's runtime cooldown | 400 invalid or unknown account id; a known account with no cooldown returns 200 with `{ ok: true, id, cleared: false }` |
 | `GET, PUT /api/codex-auth/active` | Read or select the active account | 400 invalid or missing account; 409 paused/legacy-row conflict |
 | `PUT /api/codex-auth/auto-switch` | Set the quota threshold for automatic account switching | 400 invalid threshold |
 | `PUT, PATCH /api/codex-auth/pool-strategy` | Update Codex account-pool selection strategy | 400 invalid strategy/config |
diff --git a/docs-site/src/content/docs/ru/reference/cli/providers-accounts.md b/docs-site/src/content/docs/ru/reference/cli/providers-accounts.md
index c1a27cb057..5ee51191e1 100644
--- a/docs-site/src/content/docs/ru/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/ru/reference/cli/providers-accounts.md
@@ -89,19 +89,26 @@ ocx login anthropic
 Поставляемая help-surface выглядит так:
 
 ```text
-Usage: ocx account <list|current|use|refresh|auto-switch|priority|login|reauth|code|cancel|remove|add-key|reset-credits> ...
-
-list [provider]     Codex account pool, OAuth accounts and API keys (identifiers shown masked as the API returns them).
-current <provider>  Show the active account or key.
-use <provider> <id> Switch the active credential; 'main' selects the Codex App login.
-refresh <provider>  Force-refresh Codex or provider quota reports.
-auto-switch <provider> <on|off|status|threshold N>  Control the Codex pool threshold.
-priority <provider> <id|main> [first|earlier|normal|later|last|-100..100|reset]  Selection order; omit the value to read it.
-remove <provider> <id> --yes  Remove a stored account or key after an existence check.
-add-key <provider> [--label <label>]  Add a key read only from piped stdin.
-login/reauth/code/cancel  Run browser or manual-code auth from a headless shell.
-reset-credits <id|main> [--consume --yes]  Inspect or consume Codex reset credits.
-Codex pool selection applies to the next request after clearing existing affinity; in-flight requests keep their captured account.
+Usage:
+  ocx account list [provider] [--json] [--all]
+  ocx account current <provider> [--json]
+  ocx account use <provider> <account-or-key-id|main> [--json]
+  ocx account refresh <provider> [--json]
+  ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]
+  ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]
+  ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]
+  ocx account remove <provider> <account-or-key-id|main> --yes [--json]
+  ocx account clear-cooldown <provider> <account-id|main> [--json]
+  ocx account add-key <provider> [--label <label>] [--json]
+  ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]
+  ocx account login <provider> [--id <account-id>] [--reauth] [--code -] [--no-wait] [--json]
+  ocx account code <provider> [--flow <flow-id>] [--json]   (reads the code from stdin)
+  ocx account cancel <provider> [--flow <flow-id>] [--json]
+  ocx account reset-credits <account-id|main> [--consume --yes] [--json]
+  ocx account main <doctor|list|register|add|switch|recover> ...
+
+List and switch provider accounts and API-key pools (masked output only).
+'main' selects the Codex App login for the openai account pool.
 ```
 
 Все подкоманды требуют запущенного прокси; CLI сам определяет записанный runtime-port. Успешные
@@ -188,14 +195,27 @@ quota-bar'ов дашборда.
 
 ### `ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]`
 
-Управляет только пулом аккаунтов Codex `openai`. `on` ставит 80%, `off` — 0%, `status` читает
-текущее значение, а `threshold <n>` принимает целое число от 0 до 100. Для других провайдеров и
-некорректных значений команда завершается кодом 1. `--json` возвращает:
+Управляет пулом Codex `openai` и поддерживаемыми OAuth-пулами `anthropic` и `google-antigravity`.
+Для `openai` команда `on` задаёт порог 80%, а `off` — 0%. Для OAuth-пулов `on` и `off`
+переключают состояние пула, сохраняя его порог. `threshold <n>` сохраняет целое число от 0 до 100 и
+включает пул, если значение ненулевое. `status` читает текущую политику. Для неподдерживаемых
+провайдеров и некорректных значений команда завершается кодом 1. `--json` возвращает:
 
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
 
+### `ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]`
+
+Задаёт отображаемый alias существующему Codex-аккаунту, OAuth-аккаунту или API-ключу; `-` удаляет
+alias. Допускается не более 80 печатных символов. Зарезервированный логин Codex App `main`
+переименовать нельзя, поэтому `ocx account alias openai main ...` завершается с кодом 1. `--json`
+возвращает:
+
+```text
+{ ok: true, provider, id, alias: string | null }
+```
+
 ### `ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]`
 
 Читает или задаёт порядок выбора одного аккаунта пула Codex: **больше — используется раньше**,
@@ -230,7 +250,7 @@ provider-specific формы команды используйте `ocx account
 остаётся пригодным для парсинга, а завершённый login-state содержит
 `catalogRefreshPending: true` без human-readable предупреждения.
 
-### `ocx account remove <provider> <id|main> --yes [--json]`
+### `ocx account remove <provider> <account-or-key-id|main> --yes [--json]`
 
 Это защищённое неинтерактивное удаление требует `--yes`. Перед удалением оно проверяет, что id
 существует; если id отсутствует, команда завершается кодом 1 и DELETE даже не отправляется.
@@ -249,6 +269,18 @@ provider-specific формы команды используйте `ocx account
 что удаление уже сохранено; human-readable вывод печатает в stderr общую рекомендацию `ocx sync` и
 по-прежнему завершается с кодом 0. Форматы удаления OAuth-аккаунтов и API-key не меняются.
 
+### `ocx account clear-cooldown <provider> <account-id|main> [--json]`
+
+Очищает один runtime cooldown в Codex-пуле `openai` или поддерживаемых OAuth-пулах `anthropic` и
+`google-antigravity`. `main` разрешён только для `openai` и в JSON нормализуется в `__main__`.
+Некорректный провайдер или неизвестный аккаунт отклоняется API с кодом 400, а CLI завершается с кодом 1.
+Для существующего аккаунта без активного cooldown операция идемпотентно успешна: API возвращает 200,
+CLI завершается с кодом 0, а `cleared` равен `false`. `--json` возвращает строго следующую форму:
+
+```text
+{ ok: true, provider, id, cleared: boolean }
+```
+
 ### `ocx account add-key <provider> [--label <label>] [--json]`
 
 Добавить и активировать ключ для API-key-провайдера. Ключ читается только из piped/redirected
@@ -264,6 +296,25 @@ security find-generic-password -w openrouter | ocx account add-key openrouter --
 `--json` возвращает `{ ok: true, id: string | null, label?: string }` и никогда не включает сам
 ключ.
 
+### `ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]`
+
+Импортирует ограниченный по размеру экспорт аккаунтов. Версия 1 принимает только провайдер
+`google-antigravity`, формат `cockpit-tools` и ровно один источник: `--file` или `--stdin`. Inline JSON
+и лишние позиционные аргументы отклоняются до проверки их содержимого. Храните экспорт приватно, а
+после использования удалите или защитите его. JSON-вывод — приведённая ниже сводка без секретов.
+Если есть хотя бы одна ошибочная или неподдерживаемая запись, команда завершается с кодом 1, иначе — 0.
+
+```text
+{
+  totalCount,
+  importedCount,
+  updatedCount,
+  failedCount,
+  unsupportedCount,
+  results: Array<{ index, status, code }>
+}
+```
+
 ### `ocx account reset-credits <id|main> [--consume --yes]`
 
 Проверить reset-credit'ы Codex для аккаунта. Расходование кредита разрушительно и требует сразу
diff --git a/docs-site/src/content/docs/ru/reference/configuration/providers.md b/docs-site/src/content/docs/ru/reference/configuration/providers.md
index c415517074..6c267de567 100644
--- a/docs-site/src/content/docs/ru/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ru/reference/configuration/providers.md
@@ -85,7 +85,6 @@ cross-route credential fallback не существует. Строки API GPT-
 | `modelContextWindows?` | `Record<string, number>` | Значения и cap'ы контекста по отдельным моделям. Перекрывают `contextWindow`: если окно неизвестно, берётся заданное значение, а более маленькая live-metadata остаётся авторитетной. |
 | `modelInputModalities?` | `Record<string, string[]>` | Подсказки modality по модели, например `["text"]` или `["text", "image"]`. |
 | `modelMaxInputTokens?` | `Record<string, number>` | Положительные лимиты max input по моделям, используемые для подсказок auto-compaction в каталоге. |
-| `modelAutoCompactTokenLimits?` | `Record<string, number>` | Мягкие бюджеты автосжатия по моделям в виде положительных безопасных целых чисел. Они могут только уменьшать эффективную границу в 90 % контекста или максимального ввода и не выдаются, если авторитетное окно контекста неизвестно. Для канонического `openai` ключами могут быть только точные поддерживаемые ID нативных моделей без префиксов провайдера или селектора аккаунта. PATCH провайдера объединяет записи: `null` для ключа удаляет его, а `null` для всего поля очищает карту. Такие маркеры `null` допустимы только в PATCH. |
 | `defaultMaxOutputTokens?` | `number` | Provider-wide fallback для `openai-chat`, когда клиент не передал `max_output_tokens`. |
 | `modelMaxOutputTokens?` | `Record<string, number>` | Положительные fallback-budget'ы `openai-chat` по моделям; exact/pattern-match имеет приоритет над provider-default. |
 | `modelCosts?` | `Record<string, Cost4>` | Отображаемые цены по моделям (USD за 1M токенов), ключ — точный upstream id модели этого провайдера (не идентификатор провайдера и не маршрутизируемая метка `provider/model`), значение — четыре поля: `input`, `output`, `cacheRead`, `cacheWrite` (пример: `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`). Любой id допустим — кастомный провайдер может указывать на любой OpenAI-совместимый endpoint через адаптер `openai-chat`, а локальные и внутренние провайдеры работают даже без строки во встроенных каталогах. Пользовательские цены имеют приоритет над встроенными каталогами в оценках `~$` в Logs и Usage; исторические записи пересчитываются по текущему оверлею, поэтому изменение цены может сдвинуть прошлые суммы (порядок: пользователь → каталог jawcode → expected-price overlay → вендорская цена модели); полностью нулевая запись переходит к следующему источнику. Каждая ставка должна быть неотрицательным конечным числом не более 1 000 000 (USD за 1M токенов); строки вне диапазона отклоняются на управляющей границе и отбрасываются при загрузке. Только оценка для отображения: оверлеи не влияют на маршрутизацию, выбор аккаунта, квоты или биллинг. |
@@ -210,6 +209,39 @@ backoff и может переключить аккаунт уже внутри
 `ocx account use anthropic <id>`.
 :::
 
+### `googleAntigravityAccountPool` (экспериментальная функция)
+
+Эта opt-in функция объединяет Google Antigravity OAuth-аккаунты только для Cloud Code Assist
+провайдера `google-antigravity`. По умолчанию она выключена. Пул никогда не передаёт credentials в
+Google AI Studio, Vertex AI, другую запись провайдера или маршрут с API-ключом.
+
+| Ключ | Тип | По умолчанию | Описание |
+| --- | --- | --- | --- |
+| `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | Включить локальную для процесса привязку сессий и ограниченный failover после финального 429 или дошедшего до прокси 402. |
+| `googleAntigravityAccountPool.autoSwitchThreshold?` | `number` | `80` | Порог расходования от 0 до 100. `quota` и `fill-first` берут максимальное cached usage, относящееся к семейству `Gem` или `Cla` запрошенной модели; при неизвестном usage сохраняется здоровый активный аккаунт. `0` отключает переключение по usage, но не восстановление после сбоев. |
+| `googleAntigravityAccountPool.strategy?` | `"quota" \| "round-robin" \| "fill-first"` | `"quota"` | Стратегия для новых или ещё не привязанных сессий. |
+| `googleAntigravityAccountPool.stickyLimit?` | `number` | `1` | Число привязок новых сессий, удерживаемых на одном выборе `round-robin`. Диапазон 1–100; на другие стратегии не влияет. |
+
+Все стратегии сохраняют существующую привязку сессии, пока аккаунт остаётся подходящим. Для новой
+или непривязанной сессии `quota` сохраняет здоровый активный аккаунт, пока относящееся к модели usage
+неизвестно или ниже порога, а затем выбирает подходящий аккаунт с наименьшим известным usage.
+`round-robin` равномерно распределяет привязки и не использует порог при обычной ротации. `fill-first`
+заполняет активный аккаунт до cooldown, недоступности или порога расходования, затем переходит к
+следующему подходящему аккаунту.
+
+Для каждого dispatch разрешается один привязанный к поколению OAuth snapshot, содержащий bearer token
+и Cloud Code Assist `projectId`. Финальный 429 или дошедший до прокси 402 переводит сбойный аккаунт в
+cooldown, очищает его привязку и пересобирает запрос для следующего подходящего snapshot. Если пригодного
+`Retry-After` нет, cooldown по умолчанию равен 60 секундам; распознанное значение upstream ограничено
+15 минутами. Один запрос допускает не более трёх failover, то есть всего четырёх upstream dispatch.
+Если все подходящие аккаунты в cooldown, клиент получает 429 с самым ранним известным `Retry-After`.
+
+:::caution[Операционная устойчивость, а не обход квот]
+Используйте пул для восстановления после временных сбоев аккаунта, а не для обхода квот или enforcement
+провайдера. Автоматизация нескольких аккаунтов может нарушать условия провайдера; оставьте функцию
+выключенной, если вы не принимаете этот риск.
+:::
+
 ### Формы управляемых записей
 
 В `apiKeys[]` записи содержат `id`, `name`, сгенерированный `key` и ISO-строки `createdAt`.
diff --git a/docs-site/src/content/docs/ru/reference/management-api.md b/docs-site/src/content/docs/ru/reference/management-api.md
index 860fb742d6..0db883fba7 100644
--- a/docs-site/src/content/docs/ru/reference/management-api.md
+++ b/docs-site/src/content/docs/ru/reference/management-api.md
@@ -168,14 +168,35 @@ Endpoint'ы storage cleanup могут перемещать или навсег
 | `POST /api/oauth/logout` | Удалить сохранённый credential выбранного провайдера | 400 unknown provider; `oauth_mutation_busy` |
 | `GET, DELETE /api/oauth/accounts` | Показать список masked-аккаунтов или удалить один аккаунт | 400 invalid provider/id; 404 account missing; `oauth_mutation_busy` |
 | `PUT /api/oauth/accounts/active` | Выбрать активный OAuth-аккаунт | 400 invalid provider/account; `oauth_mutation_busy` |
-| `GET, PUT, PATCH /api/oauth/accounts/pool` | Прочитать или обновить policy Anthropic OAuth pool | 400 non-Anthropic provider or invalid policy |
-| `POST /api/oauth/accounts/clear-cooldown` | Очистить runtime cooldown одного OAuth-аккаунта | 400 invalid provider/account |
+| `GET, PUT, PATCH /api/oauth/accounts/pool` | Прочитать или обновить policy Anthropic или Google Antigravity OAuth pool | 400 unsupported provider or invalid policy |
+| `POST /api/oauth/accounts/clear-cooldown` | Очистить runtime cooldown одного OAuth-аккаунта | 400 некорректный provider или неизвестный аккаунт; известный аккаунт без cooldown возвращает 200 и `{ ok: true, cleared: false }` |
 | `PUT /api/oauth/accounts/alias` | Задать или очистить alias OAuth-аккаунта | 400 invalid provider/account/alias |
 | `GET, POST, DELETE /api/providers/keys` | Показать список masked provider-key'ов, добавить/активировать один или удалить один | 400 invalid input; 404 provider/key missing |
 | `PUT /api/providers/keys/active` | Выбрать активный ключ провайдера | 400 invalid input; 404 provider/key missing |
 | `PUT /api/providers/keys/alias` | Задать или очистить alias provider-key'а | 400 invalid input; 404 provider/key missing |
 | `GET, POST, PATCH, DELETE /api/keys` | Показать список, создать, отредактировать или удалить admission key data plane | 400 invalid body/id; 404 key missing |
 
+#### Контракт OAuth-пула в рамках провайдера
+
+`GET /api/oauth/accounts/pool?provider=anthropic|google-antigravity` требует query-параметр `provider`.
+Верхнеуровневые поля ответа: `provider`, `enabled`, `autoSwitchThreshold` (по умолчанию `80`),
+`strategy` (нормализуется в `quota`, `round-robin` или `fill-first`), `stickyLimit` (нормализуется в
+целое число от 1 до 100) и `experimental: true`.
+
+`PUT` и `PATCH /api/oauth/accounts/pool` требуют JSON-объект с `provider` и принимают необязательные
+`enabled` (boolean), `autoSwitchThreshold` (целое число 0–100), `strategy` (`quota`, `round-robin` или
+`fill-first`) и `stickyLimit` (целое число 1–100). Пропущенные поля сохраняют текущее значение или
+значение по умолчанию. Успешный ответ содержит на верхнем уровне `{ ok: true, provider, enabled,
+autoSwitchThreshold, strategy, stickyLimit, experimental: true }`; обёртки `config` нет. Некорректный
+или не являющийся объектом body, неподдерживаемый provider или некорректное поле возвращает обычный
+ответ с ошибкой HTTP 400.
+
+`POST /api/oauth/accounts/clear-cooldown` принимает `{ provider, accountId }`. Provider должен быть
+`anthropic` или `google-antigravity`, а аккаунт должен существовать внутри этого provider. Неизвестный
+аккаунт возвращает обычную ошибку HTTP 400 `account not found`. Известный аккаунт без cooldown даёт
+идемпотентный успех: HTTP 200 с `{ ok: true, cleared: false }`. Эти ответы не содержат token'ов,
+email-адресов или полных key.
+
 Ответы со списками credential'ов намеренно маскируются. OAuth access-token'ы и полные API-key'и
 провайдеров клиентам дашборда не возвращаются.
 
@@ -234,7 +255,7 @@ picker изменилась. `catalogRefreshPending: true` в успешном 
 | `PUT /api/codex-auth/accounts/alias` | Задать или очистить alias аккаунта | 400 invalid account/alias |
 | `PUT /api/codex-auth/accounts/pause` | Поставить один аккаунт на паузу или снять её | 400 invalid account/state; 404 missing account |
 | `PUT /api/codex-auth/accounts/pause-exhausted` | Поставить на паузу аккаунты с исчерпанной квотой | Сбои mutation-lock превращаются в 503 |
-| `POST /api/codex-auth/accounts/clear-cooldown` | Очистить runtime cooldown для одного аккаунта или для всех | 400 invalid id |
+| `POST /api/codex-auth/accounts/clear-cooldown` | Очистить runtime cooldown одного аккаунта | 400 некорректный или неизвестный id аккаунта; известный аккаунт без cooldown возвращает 200 и `{ ok: true, id, cleared: false }` |
 | `GET, PUT /api/codex-auth/active` | Прочитать или выбрать активный аккаунт | 400 invalid or missing account; 409 paused/legacy-row conflict |
 | `PUT /api/codex-auth/auto-switch` | Задать порог квоты для автоматического переключения аккаунтов | 400 invalid threshold |
 | `PUT, PATCH /api/codex-auth/pool-strategy` | Обновить стратегию выбора в пуле аккаунтов Codex | 400 invalid strategy/config |
diff --git a/docs-site/src/content/docs/tr/reference/cli/providers-accounts.md b/docs-site/src/content/docs/tr/reference/cli/providers-accounts.md
index e8018aeeab..b20f062b3e 100644
--- a/docs-site/src/content/docs/tr/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/tr/reference/cli/providers-accounts.md
@@ -108,20 +108,26 @@ Bir sağlayıcı için saklanan OAuth kimlik bilgisini kaldırın.
 listeleyin ve değiştirin. Sağlanan yardım arayüzü şöyledir:
 
 ```text
-Usage: ocx account <list|current|use|refresh|auto-switch|priority|login|reauth|code|cancel|remove|add-key|reset-credits> ...
-
-list [provider]     Codex account pool, OAuth accounts and API keys (identifiers shown masked as the API returns them).
-current <provider>  Show the active account or key.
-use <provider> <id> Switch the active credential; 'main' selects the Codex App login.
-refresh <provider>  Force-refresh Codex or provider quota reports.
-auto-switch <provider> <on|off|status|threshold N>  Control the Codex pool threshold.
-priority <provider> <id|main> [first|earlier|normal|later|last|-100..100|reset]  Selection order; omit the value to read it.
-remove <provider> <id> --yes  Remove a stored account or key after an existence check.
-add-key <provider> [--label <label>]  Add a key read only from piped stdin.
-login/reauth/code/cancel  Run browser or manual-code auth from a headless shell.
-reset-credits <id|main> [--consume --yes]  Inspect or consume Codex reset credits.
-Switching the active account takes effect immediately; running threads move on their next request, and in-flight requests keep the account they captured.
-A selection-order change applies from the next unbound request and never moves a bound thread.
+Usage:
+  ocx account list [provider] [--json] [--all]
+  ocx account current <provider> [--json]
+  ocx account use <provider> <account-or-key-id|main> [--json]
+  ocx account refresh <provider> [--json]
+  ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]
+  ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]
+  ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]
+  ocx account remove <provider> <account-or-key-id|main> --yes [--json]
+  ocx account clear-cooldown <provider> <account-id|main> [--json]
+  ocx account add-key <provider> [--label <label>] [--json]
+  ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]
+  ocx account login <provider> [--id <account-id>] [--reauth] [--code -] [--no-wait] [--json]
+  ocx account code <provider> [--flow <flow-id>] [--json]   (reads the code from stdin)
+  ocx account cancel <provider> [--flow <flow-id>] [--json]
+  ocx account reset-credits <account-id|main> [--consume --yes] [--json]
+  ocx account main <doctor|list|register|add|switch|recover> ...
+
+List and switch provider accounts and API-key pools (masked output only).
+'main' selects the Codex App login for the openai account pool.
 ```
 
 Tüm alt komutlar proxy'nin çalışmasını gerektirir; CLI kaydedilen çalışma zamanı
@@ -229,15 +235,28 @@ eşleşen null veya eski bir rapora düşer (çıkış 0).
 
 ### `ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]`
 
-Yalnızca `openai` Codex hesap havuzunu denetler. `on` %80'i ayarlar, `off` %0'ı
-ayarlar, `status` geçerli değeri okur ve `threshold <n>` 0 ile 100 arasında bir
-tamsayı kabul eder. Diğer sağlayıcılar ve geçersiz değerler 1 ile çıkar.
-`--json` şunu döndürür:
+`openai` Codex hesap havuzunu ve desteklenen `anthropic` / `google-antigravity` OAuth
+havuzlarını denetler. `openai` için `on` eşiği %80'e, `off` ise %0'a ayarlar. OAuth
+havuzlarında `on` / `off`, mevcut eşiği koruyarak havuzun etkin durumunu değiştirir.
+`threshold <n>`, 0 ile 100 arasındaki bir tamsayıyı kaydeder ve değer sıfır değilse
+havuzu etkinleştirir. `status` geçerli politikayı okur. Desteklenmeyen sağlayıcılar ve
+geçersiz değerler 1 ile çıkar. `--json` şunu döndürür:
 
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
 
+### `ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]`
+
+Mevcut bir Codex hesabına, OAuth hesabına veya API anahtarına görüntüleme takma adı verir; temizlemek
+için `-` kullanın. Takma ad en fazla 80 yazdırılabilir karakter olabilir. Ayrılmış Codex App girişi
+`main` yeniden adlandırılamaz; bu nedenle `ocx account alias openai main ...` 1 ile çıkar. `--json`
+şunu döndürür:
+
+```text
+{ ok: true, provider, id, alias: string | null }
+```
+
 ### `ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]`
 
 Bir Codex havuz hesabının seçim sırasını okur veya ayarlar: **daha yüksek olan
@@ -284,7 +303,7 @@ sync` kurtarma rehberliği yazdırır. `--json`, stdout'u ayrıştırılabilir t
 insan uyarısı olmadan tamamlanan giriş durumunda `catalogRefreshPending: true`
 taşır.
 
-### `ocx account remove <provider> <id|main> --yes [--json]`
+### `ocx account remove <provider> <account-or-key-id|main> --yes [--json]`
 
 Bu korumalı, etkileşimsiz silme işlemi `--yes` gerektirir. Silmeden önce
 kimliğin var olduğunu doğrular; eksik bir kimlik DELETE göndermeden 1 ile çıkar.
@@ -305,6 +324,18 @@ hesap silme işlemi zaten kaydedilmiştir; insan çıktısı stderr'e genel `ocx
 kurtarma rehberliği yazdırır ve yine de 0 ile çıkar. OAuth hesabı ve API
 anahtarı kaldırma zarfları bu alanı kazanmaz.
 
+### `ocx account clear-cooldown <provider> <account-id|main> [--json]`
+
+`openai` Codex havuzunda veya desteklenen `anthropic` ve `google-antigravity` OAuth havuzlarında bir
+çalışma zamanı soğumasını temizler. `main` yalnızca `openai` için kabul edilir ve JSON'da `__main__`
+olarak normalleştirilir. Geçersiz sağlayıcı veya bilinmeyen hesap API tarafından 400 ile reddedilir ve
+CLI 1 ile çıkar. Etkin soğuması olmayan gerçek bir hesap için işlem idempotent olarak başarılıdır: API
+200 döndürür, CLI 0 ile çıkar ve `cleared` değeri `false` olur. `--json` tam olarak şunu döndürür:
+
+```text
+{ ok: true, provider, id, cleared: boolean }
+```
+
 ### `ocx account add-key <provider> [--label <label>] [--json]`
 
 Bir API anahtarı sağlayıcısı için bir anahtar ekler ve etkinleştirir. Anahtar
@@ -321,6 +352,25 @@ security find-generic-password -w openrouter | ocx account add-key openrouter --
 `--json`, `{ ok: true, id: string | null, label?: string }` döndürür ve asla
 anahtarı içermez.
 
+### `ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]`
+
+Boyutu sınırlı bir hesap dışa aktarımını içe aktarır. Sürüm 1 yalnızca `google-antigravity`
+sağlayıcısını ve `cockpit-tools` biçimini, `--file` ya da `--stdin` kaynaklarından tam olarak biriyle
+kabul eder. Satır içi JSON ve ek konumsal argümanlar içerikleri incelenmeden reddedilir. Dışa aktarımları
+gizli tutun; kullandıktan sonra silin veya güvenli biçimde saklayın. JSON çıktısı aşağıdaki gizli bilgi
+içermeyen özettir. Başarısız veya desteklenmeyen kayıt varsa komut 1, yoksa 0 ile çıkar.
+
+```text
+{
+  totalCount,
+  importedCount,
+  updatedCount,
+  failedCount,
+  unsupportedCount,
+  results: Array<{ index, status, code }>
+}
+```
+
 ### `ocx account reset-credits <id|main> [--consume --yes]`
 
 Bir hesap için Codex sıfırlama kredilerini inceleyin. Bir krediyi tüketmek
@@ -450,4 +500,3 @@ kataloğu reddeder, bu nedenle `add`, `edit` ve yönetim API'si katalog
 yazıcısının daha sonra çıkarması gereken bir şeyi saklamak yerine hatalı değeri
 reddeder (#759).
 
-
diff --git a/docs-site/src/content/docs/tr/reference/configuration/providers.md b/docs-site/src/content/docs/tr/reference/configuration/providers.md
index 4db3211bc5..7f0ae6628f 100644
--- a/docs-site/src/content/docs/tr/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/tr/reference/configuration/providers.md
@@ -91,7 +91,6 @@ alanlı seçilmiş kimlikleri yalın kimliklere yeniden yazar.
 | `modelContextWindows?` | `Record<string, number>` | Model başına bağlam geri dönüşleri/sınırları. Bunlar `contextWindow`'u geçersiz kılar: bilinmeyen bir pencere yapılandırılmış değeri kullanırken, daha küçük canlı meta veriler yetkili kalır. |
 | `modelInputModalities?` | `Record<string, string[]>` | Model başına girdi ipuçları, örn. `["text"]` veya `["text", "image"]`. |
 | `modelMaxInputTokens?` | `Record<string, number>` | Katalog otomatik sıkıştırma ipuçları için kullanılan pozitif model başına maksimum girdi sınırları. |
-| `modelAutoCompactTokenLimits?` | `Record<string, number>` | Model başına pozitif güvenli tamsayı biçiminde yumuşak otomatik sıkıştırma bütçeleri. Değerler yalnızca bağlamın veya maksimum girdinin etkin %90 zarfını düşürebilir ve yetkili bir bağlam penceresi bilinmiyorsa yayımlanmaz. Canonical `openai` için anahtarlar, sağlayıcı veya hesap seçici öneki olmadan desteklenen tam yerel model kimlikleri olmalıdır. Sağlayıcı PATCH girdileri birleştirir; bir anahtarı `null` yapmak o anahtarı siler, alanın tamamını `null` yapmak haritayı temizler. Bu `null` silme işaretleri yalnızca PATCH içindir. |
 | `defaultMaxOutputTokens?` | `number` | İstemci `max_output_tokens` değerini atladığında sağlayıcı genelinde `openai-chat` geri dönüşü. |
 | `modelMaxOutputTokens?` | `Record<string, number>` | Pozitif model başına `openai-chat` geri dönüş bütçeleri; tam/kalıp eşleşmeleri sağlayıcı varsayılanını yener. |
 | `modelCosts?` | `Record<string, Cost4>` | Sağlayıcının tam yukarı akış model kimliğine göre anahtarlanan model başına görüntüleme fiyatları (1M token başına USD) — bir sağlayıcı tanımlayıcısı veya yönlendirilen `provider/model` etiketi değil, örn. `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`. Herhangi bir model kimliği geçerli bir anahtardır — özel sağlayıcılar `openai-chat` adaptörü aracılığıyla herhangi bir OpenAI uyumlu uç noktayı hedefleyebilir ve yerel veya dahili sağlayıcı kimlikleri yerleşik kataloglarda bulunmasalar bile çalışır. Kullanıcı tarafından yapılandırılan fiyatlar Günlükler `~$` ve Kullanım tahminlerinde yerleşik katalogları yener; geçmiş girdiler geçerli katmandan yeniden fiyatlandırılır, bu nedenle bir fiyatı düzenlemek geçmiş toplamları değiştirebilir. Geri dönüş sırası: kullanıcı `modelCosts` → jawcode kataloğu → beklenen fiyat katmanı → model düzeyinde satıcı geri dönüşü ve tamamen sıfır bir girdi bu dizideki bir sonraki kaynağa düşer. Her oran en fazla 1.000.000 (1M token başına USD) olan negatif olmayan sonlu bir sayı olmalıdır; aralık dışı satırlar yönetim sınırı tarafından reddedilir ve yükleme sırasında bırakılır. Yalnızca görüntüleme zamanı tahmini: katmanlar yönlendirmeyi, hesap seçimini, kotaları veya faturalandırmayı asla etkilemez. |
@@ -234,6 +233,38 @@ Anthropic hesap politikası riskini anlamadığınız sürece bunu devre dışı
 Emin olmadığınızda manuel `ocx account use anthropic <id>` geçişini tercih edin.
 :::
 
+### `googleAntigravityAccountPool` (deneysel)
+
+Bu isteğe bağlı özellik Google Antigravity OAuth hesaplarını yalnızca `google-antigravity` Cloud Code
+Assist sağlayıcısı için havuzlar. Varsayılan olarak devre dışıdır. Havuz, kimlik bilgilerini hiçbir
+zaman Google AI Studio, Vertex AI, başka bir sağlayıcı kaydı veya API anahtarı rotasına vermez.
+
+| Anahtar | Tip | Varsayılan | Açıklama |
+| --- | --- | --- | --- |
+| `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | İşleme özgü oturum bağlılığını ve son 429 ya da yüzeye çıkan 402 yanıtında sınırlı yük devretmeyi etkinleştirir. |
+| `googleAntigravityAccountPool.autoSwitchThreshold?` | `number` | `80` | 0–100 kullanım tüketme eşiği. `quota` ve `fill-first`, istek modelinin `Gem` veya `Cla` ailesiyle ilgili önbelleğe alınmış en yüksek kullanımı kullanır; kullanım bilinmiyorsa sağlıklı etkin hesap korunur. `0` yalnızca kullanıma dayalı geçişi kapatır, hata kurtarmayı kapatmaz. |
+| `googleAntigravityAccountPool.strategy?` | `"quota" \| "round-robin" \| "fill-first"` | `"quota"` | Yeni veya henüz bağlanmamış oturumların stratejisi. |
+| `googleAntigravityAccountPool.stickyLimit?` | `number` | `1` | Bir `round-robin` seçiminde tutulan yeni oturum bağlama sayısı. Aralık 1–100'dür; diğer stratejileri etkilemez. |
+
+Tüm stratejiler, hesap uygun kaldığı sürece mevcut oturum bağlılığını korur. Yeni veya bağlanmamış bir
+oturumda `quota`, ilgili kullanım bilinmiyorsa ya da eşik altındaysa sağlıklı etkin hesabı korur; ardından
+bilinen kullanımı en düşük uygun hesabı seçer. `round-robin` bağlamaları eşit dağıtır ve normal rotasyonda
+eşiği kullanmaz. `fill-first`, etkin hesabı soğuma, kullanılamazlık veya tüketme eşiğine kadar doldurur,
+sonra sıradaki uygun hesaba geçer.
+
+Her gönderim bearer token ile Cloud Code Assist `projectId` değerini birlikte içeren, üretime bağlı tek
+bir OAuth anlık görüntüsü çözümler. Son 429 veya yüzeye çıkan 402, başarısız hesabı soğumaya alır,
+bağlılığını temizler ve isteği başka bir uygun anlık görüntü için yeniden kurar. Kullanılabilir bir
+`Retry-After` yoksa varsayılan soğuma 60 saniyedir; ayrıştırılmış upstream değeri en fazla 15 dakikadır.
+Bir istek en fazla üç yük devretmeye, yani toplam dört upstream gönderime izin verir. Uygun hesapların
+tamamı soğuyorsa istemci en erken bilinen `Retry-After` ile 429 alır.
+
+:::caution[Operasyonel dayanıklılık; kota aşma yöntemi değildir]
+Bu havuzu geçici hesap hatalarından kurtulmak için kullanın; kota veya sağlayıcı yaptırımını aşmak için
+kullanmayın. Çoklu hesap otomasyonu sağlayıcı koşullarını ihlal edebilir; bu riski kabul etmiyorsanız
+özelliği devre dışı bırakın.
+:::
+
 ### Yönetilen kayıt biçimleri
 
 `apiKeys[]` girdileri `id`, `name`, oluşturulan `key` ve ISO `createdAt`
diff --git a/docs-site/src/content/docs/tr/reference/management-api.md b/docs-site/src/content/docs/tr/reference/management-api.md
index d4a0ed290b..a53f2c451a 100644
--- a/docs-site/src/content/docs/tr/reference/management-api.md
+++ b/docs-site/src/content/docs/tr/reference/management-api.md
@@ -185,14 +185,35 @@ gönderin. Kurtarma gerekebileceğinde karantinayı tercih edin.
 | `POST /api/oauth/logout` | Seçilen sağlayıcı kimlik bilgisini kaldırın | 400 bilinmeyen sağlayıcı; `oauth_mutation_busy` |
 | `GET, DELETE /api/oauth/accounts` | Maskelenmiş hesapları listeleyin veya bir hesabı kaldırın | 400 geçersiz sağlayıcı/kimlik; 404 hesap eksik; `oauth_mutation_busy` |
 | `PUT /api/oauth/accounts/active` | Aktif OAuth hesabını seçin | 400 geçersiz sağlayıcı/hesap; `oauth_mutation_busy` |
-| `GET, PUT, PATCH /api/oauth/accounts/pool` | Anthropic OAuth havuz politikasını okuyun veya güncelleyin | 400 Anthropic olmayan sağlayıcı veya geçersiz politika |
-| `POST /api/oauth/accounts/clear-cooldown` | Bir OAuth hesabının çalışma zamanı soğuma süresini temizleyin | 400 geçersiz sağlayıcı/hesap |
+| `GET, PUT, PATCH /api/oauth/accounts/pool` | Anthropic veya Google Antigravity OAuth havuz politikasını okuyun veya güncelleyin | 400 desteklenmeyen sağlayıcı veya geçersiz politika |
+| `POST /api/oauth/accounts/clear-cooldown` | Bir OAuth hesabının çalışma zamanı soğumasını temizleyin | 400 geçersiz sağlayıcı veya bilinmeyen hesap; soğuması olmayan bilinen hesap 200 ve `{ ok: true, cleared: false }` döndürür |
 | `PUT /api/oauth/accounts/alias` | Bir OAuth hesap takma adını ayarlayın veya temizleyin | 400 geçersiz sağlayıcı/hesap/takma ad |
 | `GET, POST, DELETE /api/providers/keys` | Maskelenmiş sağlayıcı anahtarlarını listeleyin, bir tane ekleyin/etkinleştirin veya kaldırın | 400 geçersiz girdi; 404 sağlayıcı/anahtar eksik |
 | `PUT /api/providers/keys/active` | Bir sağlayıcının etkin anahtarını seçin | 400 geçersiz girdi; 404 sağlayıcı/anahtar eksik |
 | `PUT /api/providers/keys/alias` | Bir sağlayıcı anahtarı takma adını ayarlayın veya temizleyin | 400 geçersiz girdi; 404 sağlayıcı/anahtar eksik |
 | `GET, POST, PATCH, DELETE /api/keys` | Veri düzlemi kabul anahtarlarını listeleyin, oluşturun, düzenleyin veya silin | 400 geçersiz gövde/kimlik; 404 anahtar eksik |
 
+#### Sağlayıcı kapsamlı OAuth havuzu sözleşmesi
+
+`GET /api/oauth/accounts/pool?provider=anthropic|google-antigravity`, `provider` sorgu parametresini
+zorunlu kılar. Yanıtın üst düzey alanları `provider`, `enabled`, `autoSwitchThreshold` (varsayılan
+`80`), `strategy` (`quota`, `round-robin` veya `fill-first` olarak normalleştirilir), `stickyLimit`
+(1–100 aralığında bir tamsayı olarak normalleştirilir) ve `experimental: true` değerleridir.
+
+`PUT` ve `PATCH /api/oauth/accounts/pool`, `provider` içeren bir JSON nesnesi gerektirir ve isteğe bağlı
+`enabled` (boolean), `autoSwitchThreshold` (0–100 tamsayısı), `strategy` (`quota`, `round-robin` veya
+`fill-first`) ve `stickyLimit` (1–100 tamsayısı) alanlarını kabul eder. Atlanan alanlar geçerli veya
+varsayılan değerlerini korur. Başarı yanıtı üst düzeyde `{ ok: true, provider, enabled,
+autoSwitchThreshold, strategy, stickyLimit, experimental: true }` döndürür; `config` sarmalayıcısı
+yoktur. Bozuk veya nesne olmayan gövde, desteklenmeyen provider ya da geçersiz alan genel bir HTTP
+400 hata yanıtı döndürür.
+
+`POST /api/oauth/accounts/clear-cooldown`, `{ provider, accountId }` gövdesini alır. Provider
+`anthropic` veya `google-antigravity` olmalı ve account ilgili provider içinde bulunmalıdır. Bilinmeyen
+account, genel HTTP 400 `account not found` hatası döndürür. Cooldown durumunda olmayan bilinen bir
+account için işlem idempotent olarak başarılıdır: HTTP 200 ve `{ ok: true, cleared: false }`. Bu
+yanıtlar token, e-posta adresi veya eksiksiz key içermez.
+
 Kimlik bilgisi listesi yanıtları kasıtlı olarak maskelenir. OAuth erişim
 belirteçleri ve eksiksiz sağlayıcı API anahtarları kontrol paneli istemcilerine
 döndürülmez.
@@ -258,7 +279,7 @@ devreder. Rotaları şunlardır:
 | `PUT /api/codex-auth/accounts/alias` | Bir hesap takma adını ayarlayın veya temizleyin | 400 geçersiz hesap/takma ad |
 | `PUT /api/codex-auth/accounts/pause` | Bir hesabı duraklatın veya devam ettirin | 400 geçersiz hesap/durum; 404 eksik hesap |
 | `PUT /api/codex-auth/accounts/pause-exhausted` | Kotası tükenen hesapları duraklatın | Mutasyon kilidi arızaları 503 olur |
-| `POST /api/codex-auth/accounts/clear-cooldown` | Bir hesap veya tüm hesaplar için çalışma zamanı soğuma süresini temizleyin | 400 geçersiz kimlik |
+| `POST /api/codex-auth/accounts/clear-cooldown` | Bir hesabın çalışma zamanı soğumasını temizleyin | 400 geçersiz veya bilinmeyen hesap kimliği; soğuması olmayan bilinen hesap 200 ve `{ ok: true, id, cleared: false }` döndürür |
 | `GET, PUT /api/codex-auth/active` | Aktif hesabı okuyun veya seçin | 400 geçersiz veya eksik hesap; 409 duraklatılmış/eski satır çakışması |
 | `PUT /api/codex-auth/auto-switch` | Otomatik hesap geçişi için kota eşiğini ayarlayın | 400 geçersiz eşik |
 | `PUT, PATCH /api/codex-auth/pool-strategy` | Codex hesap havuzu seçim stratejisini güncelleyin | 400 geçersiz strateji/yapılandırma |
@@ -301,4 +322,3 @@ olduğunda veya işlem başarısız olduğunda sıfır olmayan bir sonuç dönd
 Doğrudan HTTP, yukarıdaki tam uç nokta sözleşmelerine ihtiyaç duyan
 entegrasyonlar için en yararlıdır.
 
-
diff --git a/docs-site/src/content/docs/zh-cn/reference/cli/providers-accounts.md b/docs-site/src/content/docs/zh-cn/reference/cli/providers-accounts.md
index 765c21aad6..e0e4b4e153 100644
--- a/docs-site/src/content/docs/zh-cn/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/zh-cn/reference/cli/providers-accounts.md
@@ -80,19 +80,26 @@ ocx login anthropic
 通过正在运行的代理列出并切换提供方账号和 API 密钥池。随附的帮助输出如下：
 
 ```text
-Usage: ocx account <list|current|use|refresh|auto-switch|priority|login|reauth|code|cancel|remove|add-key|reset-credits> ...
-
-list [provider]     Codex account pool, OAuth accounts and API keys (identifiers shown masked as the API returns them).
-current <provider>  Show the active account or key.
-use <provider> <id> Switch the active credential; 'main' selects the Codex App login.
-refresh <provider>  Force-refresh Codex or provider quota reports.
-auto-switch <provider> <on|off|status|threshold N>  Control the Codex pool threshold.
-priority <provider> <id|main> [first|earlier|normal|later|last|-100..100|reset]  Selection order; omit the value to read it.
-remove <provider> <id> --yes  Remove a stored account or key after an existence check.
-add-key <provider> [--label <label>]  Add a key read only from piped stdin.
-login/reauth/code/cancel  Run browser or manual-code auth from a headless shell.
-reset-credits <id|main> [--consume --yes]  Inspect or consume Codex reset credits.
-Codex pool selection applies to the next request after clearing existing affinity; in-flight requests keep their captured account.
+Usage:
+  ocx account list [provider] [--json] [--all]
+  ocx account current <provider> [--json]
+  ocx account use <provider> <account-or-key-id|main> [--json]
+  ocx account refresh <provider> [--json]
+  ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]
+  ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]
+  ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]
+  ocx account remove <provider> <account-or-key-id|main> --yes [--json]
+  ocx account clear-cooldown <provider> <account-id|main> [--json]
+  ocx account add-key <provider> [--label <label>] [--json]
+  ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]
+  ocx account login <provider> [--id <account-id>] [--reauth] [--code -] [--no-wait] [--json]
+  ocx account code <provider> [--flow <flow-id>] [--json]   (reads the code from stdin)
+  ocx account cancel <provider> [--flow <flow-id>] [--json]
+  ocx account reset-credits <account-id|main> [--consume --yes] [--json]
+  ocx account main <doctor|list|register|add|switch|recover> ...
+
+List and switch provider accounts and API-key pools (masked output only).
+'main' selects the Codex App login for the openai account pool.
 ```
 
 所有子命令都要求代理正在运行；CLI 会自动解析其记录的运行时端口。成功操作的
@@ -170,14 +177,25 @@ token，也不是简单重读账号列表。`--json` 返回
 
 ### `ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]`
 
-只控制 `openai` 的 Codex 账号池。`on` 会设为 80%，`off` 会设为 0%，`status` 会读取
-当前值，而 `threshold <n>` 接受 0 到 100 之间的整数。其他提供方和无效值都会以 1
-退出。`--json` 返回：
+控制 `openai` Codex 账号池及受支持的 `anthropic`、`google-antigravity` OAuth 池。
+对于 `openai`，`on` 将阈值设为 80%，`off` 将其设为 0%。对于 OAuth 池，`on` / `off`
+会切换池的启用状态，同时保留现有阈值。`threshold <n>` 会保存 0 到 100 的整数，且非零值
+会启用池。`status` 读取当前策略。不支持的提供方和无效值都会以 1 退出。`--json` 返回：
 
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
 
+### `ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]`
+
+为已有的 Codex 账号、OAuth 账号或 API 密钥设置显示别名；传入 `-` 可清除别名。别名最多包含
+80 个可打印字符。保留的 Codex App 登录 `main` 不能重命名，因此
+`ocx account alias openai main ...` 会以 1 退出。`--json` 返回：
+
+```text
+{ ok: true, provider, id, alias: string | null }
+```
+
 ### `ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]`
 
 读取或设置某个 Codex pool 账号的选择顺序：**数值越大越先使用**，默认值为 `0`，范围是 `-100` 到
@@ -206,7 +224,7 @@ token，也不是简单重读账号列表。`--json` 返回
 `--json` 时 stdout 保持可解析，已完成的登录状态会包含 `catalogRefreshPending: true`，且不会
 打印人类可读警告。
 
-### `ocx account remove <provider> <id|main> --yes [--json]`
+### `ocx account remove <provider> <account-or-key-id|main> --yes [--json]`
 
 这个受保护的非交互式删除需要 `--yes`。删除前，它会验证 id 是否存在；缺失的 id
 会以 1 退出，而不会发送 DELETE。主 Codex App 登录不能被移除，因此会拒绝
@@ -223,6 +241,17 @@ token，也不是简单重读账号列表。`--json` 返回
 输出会在 stderr 打印通用的 `ocx sync` 恢复指引，并仍以 0 退出。OAuth 账号和 API 密钥删除的
 响应结构不会增加此字段。
 
+### `ocx account clear-cooldown <provider> <account-id|main> [--json]`
+
+清除 `openai` Codex 池或受支持的 `anthropic`、`google-antigravity` OAuth 池中的一个运行时冷却。
+`main` 只适用于 `openai`，并会在 JSON 中规范化为 `__main__`。无效提供方或未知账号会被 API
+以 400 拒绝，CLI 以 1 退出。对于不存在活跃冷却的真实账号，操作会幂等成功：API 返回 200，
+CLI 以 0 退出，且 `cleared` 为 `false`。`--json` 严格返回以下结构：
+
+```text
+{ ok: true, provider, id, cleared: boolean }
+```
+
 ### `ocx account add-key <provider> [--label <label>] [--json]`
 
 为 API 密钥提供方添加并激活一个密钥。该密钥只会从非 TTY 的管道/重定向 stdin
@@ -236,6 +265,24 @@ security find-generic-password -w openrouter | ocx account add-key openrouter --
 
 `--json` 返回 `{ ok: true, id: string | null, label?: string }`，并且绝不会包含该密钥。
 
+### `ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]`
+
+导入有大小上限的账号导出文件。版本 1 只接受 `google-antigravity` 提供方和 `cockpit-tools`
+格式，并且必须只从 `--file` 或 `--stdin` 其中一个来源读取。内联 JSON 和额外位置参数会在
+检查其内容前被拒绝。请将导出文件保密，并在使用后删除或安全保存。JSON 输出是以下不含机密的
+摘要。只要存在失败或不支持的记录，命令就以 1 退出；否则以 0 退出。
+
+```text
+{
+  totalCount,
+  importedCount,
+  updatedCount,
+  failedCount,
+  unsupportedCount,
+  results: Array<{ index, status, code }>
+}
+```
+
 ### `ocx account reset-credits <id|main> [--consume --yes]`
 
 查看某个账号的 Codex 重置额度。消耗额度会造成破坏性影响，因此同时需要 `--consume`
diff --git a/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md b/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
index 3630a9ba6c..059157efab 100644
--- a/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
@@ -72,7 +72,6 @@ selector，而不是分配一个新名称。
 | `modelContextWindows?` | `Record<string, number>` | 按模型设置的上下文数值与上限。优先于 `contextWindow`：窗口未知时采用所配置的数值，而更小的实时元数据仍然优先。 |
 | `modelInputModalities?` | `Record<string, string[]>` | 按模型设置的输入提示，例如 `["text"]` 或 `["text", "image"]`。 |
 | `modelMaxInputTokens?` | `Record<string, number>` | 正数型、按模型设置的最大输入限制，用于目录自动压缩提示。 |
-| `modelAutoCompactTokenLimits?` | `Record<string, number>` | 按模型设置的正安全整数软自动压缩预算。该值只能降低“上下文或最大输入的 90%”这一有效上限；没有已知的权威上下文窗口时不会输出。对于规范 `openai`，键必须是受支持的精确原生模型 ID，且不得包含提供者或账户选择器前缀。提供者 PATCH 会合并条目；将某个键设为 `null` 会删除该键，将整个字段设为 `null` 会清空映射。这些 `null` 删除标记仅适用于 PATCH。 |
 | `defaultMaxOutputTokens?` | `number` | 当客户端省略 `max_output_tokens` 时，`openai-chat` 的提供者级回退值。 |
 | `modelMaxOutputTokens?` | `Record<string, number>` | 正数型、按模型设置的 `openai-chat` 回退预算；精确/模式匹配优先于提供者默认值。 |
 | `modelCosts?` | `Record<string, Cost4>` | 按模型设置的显示价格（每 100 万 token 的美元数），以该提供者的精确上游模型 ID 为键（不是提供者标识符或路由后的 `provider/model` 标签），值为四个字段：`input`、`output`、`cacheRead`、`cacheWrite`（示例：`{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`）。任何模型 ID 都是有效键——自定义提供者可以通过 `openai-chat` 适配器指向任意 OpenAI 兼容端点，即使不存在于内置目录中，本地 OpenAI 兼容和内部提供者的 ID 同样有效。用户配置的价格在 Logs 的 `~$` 和 Usage 估算中优先于内置目录；历史条目也会按当前覆盖项重新计价，因此修改价格可能改变过去的总额（回退顺序：用户配置 → jawcode 目录 → expected-price 覆盖 → 模型级厂商价格）；全零条目会回退到该顺序中的下一个来源。每个费率必须是大于等于 0 的有限数字，且不超过 1,000,000（每 100 万 token 的美元数）；超出范围的条目会在管理边界被拒绝，并在加载时被丢弃。仅用于显示的估算：覆盖项不影响路由、账户选择、配额或计费。 |
@@ -167,6 +166,34 @@ affinity。这些策略不能规避 provider enforcement。
 除非你理解 Anthropic 账户策略风险，否则请保持关闭。若不确定，优先手动使用 `ocx account use anthropic <id>` 切换。
 :::
 
+### `googleAntigravityAccountPool`（实验性）
+
+此可选功能仅为 `google-antigravity` Cloud Code Assist 提供方池化 Google Antigravity OAuth
+账户，默认关闭。该池绝不会把凭据用于 Google AI Studio、Vertex AI、其他提供方条目或 API 密钥路由。
+
+| 键 | 类型 | 默认值 | 说明 |
+| --- | --- | --- | --- |
+| `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | 启用进程本地会话亲和性，以及最终 429 或透传到此处的 402 响应触发的有界故障转移。 |
+| `googleAntigravityAccountPool.autoSwitchThreshold?` | `number` | `80` | 0–100 的使用率耗尽阈值。`quota` 和 `fill-first` 使用与请求模型 `Gem` 或 `Cla` 系列相关的最大缓存使用率；使用率未知时保留健康的活跃账户。`0` 只关闭按使用率切换，不关闭故障恢复。 |
+| `googleAntigravityAccountPool.strategy?` | `"quota" \| "round-robin" \| "fill-first"` | `"quota"` | 新会话或未绑定会话的策略。 |
+| `googleAntigravityAccountPool.stickyLimit?` | `number` | `1` | 一次 `round-robin` 选择所保留的新会话绑定数。范围 1–100；不影响其他策略。 |
+
+只要账户仍然合格，所有策略都会保留已有的会话亲和性。对于新会话或未绑定会话，`quota` 会在
+相关使用率未知或低于阈值时保留健康的活跃账户，之后选择已知使用率最低的合格账户。
+`round-robin` 均匀分配绑定，普通轮换不使用该阈值。`fill-first` 会持续使用活跃账户，直到账户
+进入冷却、不可用或达到耗尽阈值，再前进到下一个合格账户。
+
+每次派发都会解析一个绑定到该代请求的 OAuth 快照，其中同时包含 bearer token 和 Cloud Code
+Assist `projectId`。最终 429 或透传到此处的 402 会让失败账户进入冷却、清除其亲和性，并用另一个
+合格快照重建请求。如果没有可用的 `Retry-After`，默认冷却 60 秒；成功解析的上游值最高限制为
+15 分钟。单个请求最多允许 3 次故障转移，即总计 4 次上游派发。如果所有合格账户都在冷却，
+客户端会收到 429，并带上已知的最早 `Retry-After`。
+
+:::caution[用于运行韧性，而不是绕过配额]
+请只用该池从暂时性账户故障中恢复，不要用它规避配额或提供方管控。多账户自动化可能违反提供方
+条款；如果不接受这一风险，请保持关闭。
+:::
+
 ### 托管记录形状
 
 `apiKeys[]` 条目包含 `id`、`name`、生成的 `key` 以及 ISO 格式的 `createdAt` 字符串。`codexAccounts[]` 条目要求有 `id`、`email` 和 `isMain`，并可选 `plan`、`chatgptAccountId` 和具备隐私安全性的 `logLabel`。这些记录通常由仪表板管理。
diff --git a/docs-site/src/content/docs/zh-cn/reference/management-api.md b/docs-site/src/content/docs/zh-cn/reference/management-api.md
index ab06540139..6dc075c9c2 100644
--- a/docs-site/src/content/docs/zh-cn/reference/management-api.md
+++ b/docs-site/src/content/docs/zh-cn/reference/management-api.md
@@ -150,14 +150,33 @@ Authorization: Bearer <admin-token>
 | `POST /api/oauth/logout` | 移除选定的 provider 凭证 | 400 provider 未知；`oauth_mutation_busy` |
 | `GET, DELETE /api/oauth/accounts` | 列出已脱敏账户或移除一个账户 | 400 provider/id 无效；404 账户缺失；`oauth_mutation_busy` |
 | `PUT /api/oauth/accounts/active` | 选择当前活跃的 OAuth 账户 | 400 provider/账户无效；`oauth_mutation_busy` |
-| `GET, PUT, PATCH /api/oauth/accounts/pool` | 读取或更新 Anthropic OAuth 池策略 | 400 非 Anthropic provider 或策略无效 |
-| `POST /api/oauth/accounts/clear-cooldown` | 清除一个 OAuth 账户的运行时冷却 | 400 provider/账户无效 |
+| `GET, PUT, PATCH /api/oauth/accounts/pool` | 读取或更新 Anthropic 或 Google Antigravity OAuth 池策略 | 400 不支持的 provider 或策略无效 |
+| `POST /api/oauth/accounts/clear-cooldown` | 清除一个 OAuth 账户的运行时冷却 | 400：provider 无效或账户未知；已知账户没有冷却时返回 200 和 `{ ok: true, cleared: false }` |
 | `PUT /api/oauth/accounts/alias` | 设置或清除 OAuth 账户别名 | 400 provider/账户/别名无效 |
 | `GET, POST, DELETE /api/providers/keys` | 列出已脱敏的 provider 密钥，添加/激活一个，或移除一个 | 400 输入无效；404 provider/密钥缺失 |
 | `PUT /api/providers/keys/active` | 选择某个 provider 的活跃密钥 | 400 输入无效；404 provider/密钥缺失 |
 | `PUT /api/providers/keys/alias` | 设置或清除 provider 密钥别名 | 400 输入无效；404 provider/密钥缺失 |
 | `GET, POST, PATCH, DELETE /api/keys` | 列出、创建、编辑或删除数据平面准入密钥 | 400 请求体/id 无效；404 密钥缺失 |
 
+#### 提供方范围内的 OAuth 池契约
+
+`GET /api/oauth/accounts/pool?provider=anthropic|google-antigravity` 要求提供 `provider` 查询参数。
+响应的顶层字段为 `provider`、`enabled`、`autoSwitchThreshold`（默认 `80`）、`strategy`（规范化为
+`quota`、`round-robin` 或 `fill-first`）、`stickyLimit`（规范化为 1～100 的整数）以及
+`experimental: true`。
+
+`PUT` 和 `PATCH /api/oauth/accounts/pool` 要求请求体为包含 `provider` 的 JSON 对象，并可选接受
+`enabled`（布尔值）、`autoSwitchThreshold`（0～100 的整数）、`strategy`（`quota`、`round-robin`
+或 `fill-first`）以及 `stickyLimit`（1～100 的整数）。省略的字段会保留当前值或默认值。成功响应
+直接在顶层返回 `{ ok: true, provider, enabled, autoSwitchThreshold, strategy, stickyLimit,
+experimental: true }`，没有 `config` 包装层。格式错误或非对象的请求体、不支持的 provider 或
+无效字段都会返回通用 HTTP 400 错误响应。
+
+`POST /api/oauth/accounts/clear-cooldown` 的请求体为 `{ provider, accountId }`。provider 必须是
+`anthropic` 或 `google-antigravity`，且账号必须存在于该 provider 内。未知账号返回通用 HTTP 400
+`account not found` 错误。已知账号即使不在冷却中也会幂等成功：HTTP 200 和
+`{ ok: true, cleared: false }`。这些响应不会包含 token、电子邮件地址或完整密钥。
+
 凭证列表响应会刻意脱敏。OAuth 访问令牌和完整的 provider API 密钥不会返回给仪表板客户端。
 
 ### Providers
@@ -212,7 +231,7 @@ Authorization: Bearer <admin-token>
 | `PUT /api/codex-auth/accounts/alias` | 设置或清除账户别名 | 400 账户/别名无效 |
 | `PUT /api/codex-auth/accounts/pause` | 暂停或恢复一个账户 | 400 账户/状态无效；404 缺少账户 |
 | `PUT /api/codex-auth/accounts/pause-exhausted` | 暂停配额已耗尽的账户 | 变更锁失败会变成 503 |
-| `POST /api/codex-auth/accounts/clear-cooldown` | 清除一个账户或所有账户的运行时冷却 | 400 id 无效 |
+| `POST /api/codex-auth/accounts/clear-cooldown` | 清除一个账户的运行时冷却 | 400：账户 id 无效或未知；已知账户没有冷却时返回 200 和 `{ ok: true, id, cleared: false }` |
 | `GET, PUT /api/codex-auth/active` | 读取或选择当前活跃账户 | 400 账户无效或缺失；409 暂停/旧行冲突 |
 | `PUT /api/codex-auth/auto-switch` | 设置自动切换账户的配额阈值 | 400 阈值无效 |
 | `PUT, PATCH /api/codex-auth/pool-strategy` | 更新 Codex 账户池选择策略 | 400 策略/配置无效 |
diff --git a/docs-site/src/content/docs/zh-tw/reference/cli/providers-accounts.md b/docs-site/src/content/docs/zh-tw/reference/cli/providers-accounts.md
index ec2063b31b..d7c1479993 100644
--- a/docs-site/src/content/docs/zh-tw/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/zh-tw/reference/cli/providers-accounts.md
@@ -59,18 +59,26 @@ ocx login anthropic
 透過執行中的代理列出並切換供應商帳號與 API-key 池。隨附的說明介面如下：
 
 ```text
-Usage: ocx account <list|current|use|refresh|auto-switch|login|reauth|code|cancel|remove|add-key|reset-credits> ...
-
-list [provider]     Codex 帳號池、OAuth 帳號與 API 金鑰（識別碼依 API 回傳遮罩顯示）。
-current <provider>  顯示現用帳號或金鑰。
-use <provider> <id> 切換現用憑證；'main' 選擇 Codex App 登入。
-refresh <provider>  強制重新整理 Codex 或供應商配額報告。
-auto-switch <provider> <on|off|status|threshold N>  控制 Codex 池閾值。
-remove <provider> <id> --yes  在存在檢查後移除已儲存的帳號或金鑰。
-add-key <provider> [--label <label>]  僅從 piped stdin 讀取並新增金鑰。
-login/reauth/code/cancel  從無頭 shell 執行瀏覽器或手動 code 認證。
-reset-credits <id|main> [--consume --yes]  檢查或消耗 Codex reset credits。
-Codex 池選擇套用於清除既有親和性後的下一個請求；進行中的請求保留其擷取的帳號。
+Usage:
+  ocx account list [provider] [--json] [--all]
+  ocx account current <provider> [--json]
+  ocx account use <provider> <account-or-key-id|main> [--json]
+  ocx account refresh <provider> [--json]
+  ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]
+  ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]
+  ocx account priority <provider> <account-id|main> [<-100..100|first|earlier|normal|later|last|reset>] [--json]
+  ocx account remove <provider> <account-or-key-id|main> --yes [--json]
+  ocx account clear-cooldown <provider> <account-id|main> [--json]
+  ocx account add-key <provider> [--label <label>] [--json]
+  ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]
+  ocx account login <provider> [--id <account-id>] [--reauth] [--code -] [--no-wait] [--json]
+  ocx account code <provider> [--flow <flow-id>] [--json]   (reads the code from stdin)
+  ocx account cancel <provider> [--flow <flow-id>] [--json]
+  ocx account reset-credits <account-id|main> [--consume --yes] [--json]
+  ocx account main <doctor|list|register|add|switch|recover> ...
+
+List and switch provider accounts and API-key pools (masked output only).
+'main' selects the Codex App login for the openai account pool.
 ```
 
 所有子指令都需要代理正在執行；CLI 自動解析其記錄的 runtime 連接埠。成功的操作離開 0。無效用法、未知供應商或帳號／金鑰 id、不可達的代理或 API 失敗則離開 1。憑證欄位完全依管理 API 回傳的方式顯示（包含其遮罩）；原始 API 金鑰與 OAuth token 永不回傳。顯示便利性在客戶端合成，與儀表板相同：`main` 是 `openai` 帳號池中 Codex App 登入的 CLI 別名，無電子郵件的 OAuth 帳號顯示為 `Account N`，而 plan／label 欄位在 plan、遮罩電子郵件、label 與遮罩金鑰之間回退。
@@ -129,17 +137,30 @@ Codex 池選擇套用於清除既有親和性後的下一個請求；進行中
 
 ### `ocx account auto-switch <provider> <on|off|status|threshold <0-100>> [--json]`
 
-僅控制 `openai` Codex 帳號池。`on` 設為 80%，`off` 設為 0%，`status` 讀取目前值，而 `threshold <n>` 接受 0 到 100 的整數。其他供應商與無效值離開 1。`--json` 回傳：
+控制 `openai` Codex 帳號池與支援的 `anthropic`、`google-antigravity` OAuth 池。
+對於 `openai`，`on` 將閾值設為 80%，`off` 將其設為 0%。對於 OAuth 池，`on` / `off`
+會切換池的啟用狀態，同時保留現有閾值。`threshold <n>` 會儲存 0 到 100 的整數，且非零值
+會啟用池。`status` 讀取目前政策。不支援的供應商與無效值會以 1 離開。`--json` 回傳：
 
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
 
+### `ocx account alias <provider> <account-or-key-id> <display-name|-> [--json]`
+
+為既有 Codex 帳號、OAuth 帳號或 API 金鑰設定顯示別名；傳入 `-` 可清除別名。別名最多包含
+80 個可列印字元。保留的 Codex App 登入 `main` 無法重新命名，因此
+`ocx account alias openai main ...` 會以 1 離開。`--json` 回傳：
+
+```text
+{ ok: true, provider, id, alias: string | null }
+```
+
 ### `ocx account login|reauth|code|cancel ...`
 
 從無頭 shell 執行基於瀏覽器或手動 code 的帳號認證。請使用 `ocx account --help` 查看供應商專屬的指令形式。
 
-### `ocx account remove <provider> <id|main> --yes [--json]`
+### `ocx account remove <provider> <account-or-key-id|main> --yes [--json]`
 
 此受保護的非互動刪除需要 `--yes`。刪除前，它驗證 id 存在；缺失的 id 離開 1 而不發送 DELETE。主要的 Codex App 登入無法被移除，因此 `remove openai main --yes` 被拒絕。刪除後，家族會再次讀取：移除 pin 的 Codex 帳號會清除 pin 並回到自動選擇；OAuth 提升第一個剩餘帳號或回報無；API-key 池提升第一個剩餘金鑰或回報無。`--json` 成功與失敗結構為：
 
@@ -148,6 +169,17 @@ Codex 池選擇套用於清除既有親和性後的下一個請求；進行中
 { error: string } // stderr, exit 1
 ```
 
+### `ocx account clear-cooldown <provider> <account-id|main> [--json]`
+
+清除 `openai` Codex 池或支援的 `anthropic`、`google-antigravity` OAuth 池中一個 runtime 冷卻。
+`main` 只適用於 `openai`，並會在 JSON 中正規化為 `__main__`。無效供應商或未知帳號會被 API
+以 400 拒絕，CLI 以 1 離開。對於沒有作用中冷卻的真實帳號，操作會冪等成功：API 回傳 200、
+CLI 以 0 離開，且 `cleared` 為 `false`。`--json` 嚴格回傳以下結構：
+
+```text
+{ ok: true, provider, id, cleared: boolean }
+```
+
 ### `ocx account add-key <provider> [--label <label>] [--json]`
 
 為 API-key 供應商新增並啟用金鑰。金鑰僅從非 TTY piped／重新導向的 stdin 讀取；互動式 TTY 輸入、空輸入、OAuth／Codex 供應商與 API 失敗離開 1。金鑰永不回顯，即使它出現在 label 中時亦然。偏好使用密碼管理員或 here-string：
@@ -159,6 +191,24 @@ security find-generic-password -w openrouter | ocx account add-key openrouter --
 
 `--json` 回傳 `{ ok: true, id: string | null, label?: string }` 且永不包含金鑰。
 
+### `ocx account import <provider> --format <format> (--file <path>|--stdin) [--json]`
+
+匯入有大小上限的帳號 export。版本 1 只接受 `google-antigravity` 供應商與 `cockpit-tools`
+格式，且必須只從 `--file` 或 `--stdin` 其中一個來源讀取。inline JSON 與額外 positional
+argument 會在檢查內容前被拒絕。請將 export 保持私密，並在使用後刪除或安全保存。JSON 輸出是
+以下不含秘密的摘要。只要有 failed 或 unsupported 記錄，指令就以 1 離開；否則以 0 離開。
+
+```text
+{
+  totalCount,
+  importedCount,
+  updatedCount,
+  failedCount,
+  unsupportedCount,
+  results: Array<{ index, status, code }>
+}
+```
+
 ### `ocx account reset-credits <id|main> [--consume --yes]`
 
 檢查帳號的 Codex reset credits。消耗 credit 是破壞性的，需要同時提供 `--consume` 與 `--yes`。
diff --git a/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md b/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
index b0a46f49ec..c74f795431 100644
--- a/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
@@ -54,7 +54,6 @@ description: 供應商項目、認證、端點、模型目錄、配額、context
 | `modelContextWindows?` | `Record<string, number>` | Per-model context 上限。這些覆寫 `contextWindow` 且永不提高較小的即時中繼資料。 |
 | `modelInputModalities?` | `Record<string, string[]>` | Per-model 輸入提示，如 `["text"]` 或 `["text", "image"]`。 |
 | `modelMaxInputTokens?` | `Record<string, number>` | 用於目錄自動壓縮提示的正數 per-model max input 限制。 |
-| `modelAutoCompactTokenLimits?` | `Record<string, number>` | Per-model 正安全整數型 soft 自動壓縮預算。此值只能降低「context 或 max input 的 90%」這個有效上限；沒有已知的權威 context window 時不會輸出。對 canonical `openai` 而言，key 必須是受支援的精確 native model ID，且不得含 provider 或 account-selector 前綴。Provider PATCH 會合併項目；將單一 key 設為 `null` 會刪除該 key，將整個欄位設為 `null` 會清空 map。這些 `null` tombstone 僅供 PATCH 使用。 |
 | `defaultMaxOutputTokens?` | `number` | 當客戶端省略 `max_output_tokens` 時的供應商範圍 `openai-chat` 後備。 |
 | `modelMaxOutputTokens?` | `Record<string, number>` | 正數 per-model `openai-chat` 後援預算；精確／模式比對勝過供應商預設。 |
 | `headers?` | `Record<string, string>` | 額外上游標頭。Authorization、cookie、API-key 標頭、內嵌換行與無效名稱被拒絕。 |
@@ -136,6 +135,34 @@ API-key 供應商可持有字面值金鑰或環境參考。OAuth 供應商使用
 除非你了解 Anthropic 帳號政策風險，否則保持停用。不確定時偏好手動 `ocx account use anthropic <id>` 切換。
 :::
 
+### `googleAntigravityAccountPool`（實驗性）
+
+此選用功能只為 `google-antigravity` Cloud Code Assist 供應商池化 Google Antigravity OAuth
+帳號，預設停用。此池絕不會把憑證用於 Google AI Studio、Vertex AI、其他供應商項目或 API-key 路由。
+
+| Key | 型別 | 預設值 | 說明 |
+| --- | --- | --- | --- |
+| `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | 啟用行程本地的 session 親和性，以及最終 429 或傳到此處的 402 回應所觸發的有界容錯移轉。 |
+| `googleAntigravityAccountPool.autoSwitchThreshold?` | `number` | `80` | 0–100 的用量耗盡閾值。`quota` 與 `fill-first` 使用和請求模型 `Gem` 或 `Cla` 家族相關的最高快取用量；用量未知時保留健康的現用帳號。`0` 只停用依用量切換，不停用故障復原。 |
+| `googleAntigravityAccountPool.strategy?` | `"quota" \| "round-robin" \| "fill-first"` | `"quota"` | 新 session 或未綁定 session 的策略。 |
+| `googleAntigravityAccountPool.stickyLimit?` | `number` | `1` | 一次 `round-robin` 選擇所保留的新 session 綁定數。範圍 1–100；不影響其他策略。 |
+
+只要帳號仍合格，所有策略都會保留既有 session 親和性。對於新 session 或未綁定 session，
+`quota` 會在相關用量未知或低於閾值時保留健康的現用帳號，之後選擇已知用量最低的合格帳號。
+`round-robin` 平均分配綁定，正常輪換不使用該閾值。`fill-first` 會持續使用現用帳號，直到帳號
+進入冷卻、不可用或到達耗盡閾值，再前進到下一個合格帳號。
+
+每次派送都會解析一份綁定到該代請求的 OAuth 快照，其中同時包含 bearer token 與 Cloud Code
+Assist `projectId`。最終 429 或傳到此處的 402 會讓失敗帳號進入冷卻、清除其親和性，並用另一份
+合格快照重建請求。若沒有可用的 `Retry-After`，預設冷卻 60 秒；成功解析的上游值最高限制為
+15 分鐘。單一請求最多允許 3 次容錯移轉，也就是總計 4 次上游派送。若所有合格帳號都在冷卻，
+客戶端會收到 429，並附上已知最早的 `Retry-After`。
+
+:::caution[用於營運韌性，而非繞過配額]
+請只用此池從暫時性帳號故障中復原，不要用來規避配額或供應商管控。多帳號自動化可能違反供應商
+條款；若不接受這項風險，請保持停用。
+:::
+
 ### 受管記錄結構
 
 `apiKeys[]` 項目包含 `id`、`name`、生成的 `key` 與 ISO `createdAt` 字串。
diff --git a/docs-site/src/content/docs/zh-tw/reference/management-api.md b/docs-site/src/content/docs/zh-tw/reference/management-api.md
index 52d6db4d8e..38ac5841e8 100644
--- a/docs-site/src/content/docs/zh-tw/reference/management-api.md
+++ b/docs-site/src/content/docs/zh-tw/reference/management-api.md
@@ -150,14 +150,33 @@ Session 簽發在需要 data-plane 認證時停用，這包含遠端綁定。遠
 | `POST /api/oauth/logout` | 移除所選的供應商憑證 | 400 未知供應商；`oauth_mutation_busy` |
 | `GET, DELETE /api/oauth/accounts` | 列出遮罩帳號或移除一個帳號 | 400 無效供應商/id；404 帳號缺失；`oauth_mutation_busy` |
 | `PUT /api/oauth/accounts/active` | 選擇現用 OAuth 帳號 | 400 無效供應商／帳號；`oauth_mutation_busy` |
-| `GET, PUT, PATCH /api/oauth/accounts/pool` | 讀取或更新 Anthropic OAuth 池政策 | 400 非 Anthropic 供應商或無效政策 |
-| `POST /api/oauth/accounts/clear-cooldown` | 清除一個 OAuth 帳號的 runtime 冷卻 | 400 無效供應商／帳號 |
+| `GET, PUT, PATCH /api/oauth/accounts/pool` | 讀取或更新 Anthropic 或 Google Antigravity OAuth 池政策 | 400 不支援的供應商或無效政策 |
+| `POST /api/oauth/accounts/clear-cooldown` | 清除一個 OAuth 帳號的 runtime 冷卻 | 400：供應商無效或帳號未知；已知帳號沒有冷卻時回傳 200 與 `{ ok: true, cleared: false }` |
 | `PUT /api/oauth/accounts/alias` | 設定或清除 OAuth 帳號別名 | 400 無效供應商／帳號／別名 |
 | `GET, POST, DELETE /api/providers/keys` | 列出遮罩供應商金鑰、新增／啟用一個或移除一個 | 400 無效輸入；404 供應商／金鑰缺失 |
 | `PUT /api/providers/keys/active` | 選擇供應商的現用金鑰 | 400 無效輸入；404 供應商／金鑰缺失 |
 | `PUT /api/providers/keys/alias` | 設定或清除供應商金鑰別名 | 400 無效輸入；404 供應商／金鑰缺失 |
 | `GET, POST, PATCH, DELETE /api/keys` | 列出、建立、編輯或刪除 data-plane 許可金鑰 | 400 無效 body/id；404 金鑰缺失 |
 
+#### 供應商範圍內的 OAuth 池契約
+
+`GET /api/oauth/accounts/pool?provider=anthropic|google-antigravity` 要求提供 `provider` 查詢參數。
+回應的頂層欄位為 `provider`、`enabled`、`autoSwitchThreshold`（預設 `80`）、`strategy`（正規化為
+`quota`、`round-robin` 或 `fill-first`）、`stickyLimit`（正規化為 1～100 的整數）以及
+`experimental: true`。
+
+`PUT` 與 `PATCH /api/oauth/accounts/pool` 要求 body 為包含 `provider` 的 JSON 物件，並可選接受
+`enabled`（boolean）、`autoSwitchThreshold`（0～100 的整數）、`strategy`（`quota`、
+`round-robin` 或 `fill-first`）以及 `stickyLimit`（1～100 的整數）。省略的欄位會保留目前值或
+預設值。成功回應直接在頂層回傳 `{ ok: true, provider, enabled, autoSwitchThreshold, strategy,
+stickyLimit, experimental: true }`，沒有 `config` 包裝層。格式錯誤或非物件的 body、不支援的
+provider 或無效欄位都會回傳通用 HTTP 400 錯誤回應。
+
+`POST /api/oauth/accounts/clear-cooldown` 的 body 為 `{ provider, accountId }`。provider 必須是
+`anthropic` 或 `google-antigravity`，且帳號必須存在於該 provider 內。未知帳號回傳通用 HTTP 400
+`account not found` 錯誤。已知帳號即使不在冷卻中也會冪等成功：HTTP 200 與
+`{ ok: true, cleared: false }`。這些回應不會包含 token、電子郵件地址或完整金鑰。
+
 憑證清單回應被刻意遮罩。OAuth access token 與完整的供應商 API 金鑰不回傳給儀表板客戶端。
 
 ### 供應商
@@ -205,7 +224,7 @@ Session 簽發在需要 data-plane 認證時停用，這包含遠端綁定。遠
 | `PUT /api/codex-auth/accounts/alias` | 設定或清除帳號別名 | 400 無效帳號／別名 |
 | `PUT /api/codex-auth/accounts/pause` | 暫停或恢復一個帳號 | 400 無效帳號／狀態；404 缺失帳號 |
 | `PUT /api/codex-auth/accounts/pause-exhausted` | 暫停配額耗盡的帳號 | 變更鎖失敗變為 503 |
-| `POST /api/codex-auth/accounts/clear-cooldown` | 清除一個或所有帳號的 runtime 冷卻 | 400 無效 id |
+| `POST /api/codex-auth/accounts/clear-cooldown` | 清除一個帳號的 runtime 冷卻 | 400：帳號 id 無效或未知；已知帳號沒有冷卻時回傳 200 與 `{ ok: true, id, cleared: false }` |
 | `GET, PUT /api/codex-auth/active` | 讀取或選擇現用帳號 | 400 無效或缺失帳號；409 暫停／舊列衝突 |
 | `PUT /api/codex-auth/auto-switch` | 設定自動帳號切換的配額閾值 | 400 無效閾值 |
 | `PUT, PATCH /api/codex-auth/pool-strategy` | 更新 Codex 帳號池選擇策略 | 400 無效策略／設定 |
diff --git a/src/cli/account-extended.ts b/src/cli/account-extended.ts
index 952a552d79..280f8be8ec 100644
--- a/src/cli/account-extended.ts
+++ b/src/cli/account-extended.ts
@@ -345,8 +345,11 @@ export async function cmdAutoSwitch(args: string[], deps: AccountDeps): Promise<
   const action = args.shift();
   if (!name || !action) return usage();
   const classified = configAndType(deps, name);
-  if ("error" in classified || classified.type !== "codex") {
-    return usage("Error: auto-switch only applies to the openai Codex account pool");
+  if ("error" in classified) return usage(`Error: ${classified.error}`);
+  const oauthPool = classified.type === "oauth"
+    && (name === "anthropic" || name === "google-antigravity");
+  if (classified.type !== "codex" && !oauthPool) {
+    return usage(`Error: ${name} does not support account-pool auto-switch`);
   }
   let threshold: number | undefined;
   if (action === "on" && args.length === 0) threshold = 80;
@@ -358,6 +361,46 @@ export async function cmdAutoSwitch(args: string[], deps: AccountDeps): Promise<
   }
   const baseUrl = await resolveBaseUrl(deps);
   if (!baseUrl) return proxyUnreachable();
+  if (oauthPool) {
+    let enabled: boolean;
+    if (action === "status") {
+      const response = await apiJson(
+        deps,
+        baseUrl,
+        "GET",
+        `/api/oauth/accounts/pool?provider=${encodeURIComponent(name)}`,
+      );
+      if (response.status === 0) return proxyUnreachable();
+      if (
+        response.status !== 200
+        || typeof response.json.enabled !== "boolean"
+        || typeof response.json.autoSwitchThreshold !== "number"
+      ) return apiError(response.json, "failed to read auto-switch status");
+      enabled = response.json.enabled;
+      threshold = response.json.autoSwitchThreshold;
+    } else {
+      const update = action === "on"
+        ? { provider: name, enabled: true }
+        : action === "off"
+          ? { provider: name, enabled: false }
+          : { provider: name, enabled: threshold! > 0, autoSwitchThreshold: threshold };
+      const response = await apiJson(deps, baseUrl, "PUT", "/api/oauth/accounts/pool", update);
+      if (response.status === 0) return proxyUnreachable();
+      if (
+        response.status !== 200
+        || typeof response.json.enabled !== "boolean"
+        || typeof response.json.autoSwitchThreshold !== "number"
+      ) return apiError(response.json, "failed to update auto-switch");
+      enabled = response.json.enabled;
+      threshold = response.json.autoSwitchThreshold;
+    }
+    if (wantsJson) {
+      console.log(JSON.stringify({ provider: name, autoSwitchThreshold: threshold, enabled }, null, 2));
+    } else {
+      console.log(enabled ? `auto-switch: on (threshold ${threshold}%)` : "auto-switch: off");
+    }
+    return 0;
+  }
   if (action === "status") {
     const response = await apiJson(deps, baseUrl, "GET", "/api/codex-auth/active");
     if (response.status === 0) return proxyUnreachable();
@@ -566,14 +609,13 @@ export async function cmdImport(args: string[], deps: AccountDeps): Promise<numb
 }
 
 /**
- * Lift a quota cooldown on a Codex account.
+ * Lift a quota cooldown on a Codex or supported OAuth pool account.
  *
  * This is the user-facing escape from the lockout described in
  * `devlog/_plan/260726_cooldown_lockout_hardening`: injected routing makes the proxy the
  * only model path for Codex Desktop, so a stuck cooldown reads as "the whole app is dead".
  *
- * Codex accounts only. API-key pools already reset their own 429 cooldowns through key
- * management (`clearKeyCooldowns`), and OAuth providers have no equivalent state here.
+ * API-key pools already reset their own 429 cooldowns through key management.
  */
 export async function cmdClearCooldown(args: string[], deps: AccountDeps): Promise<number> {
   const wantsJson = flag(args, "--json");
@@ -582,13 +624,23 @@ export async function cmdClearCooldown(args: string[], deps: AccountDeps): Promi
   if (!name || !requestedId || args.length) return usage();
   const classified = configAndType(deps, name);
   if ("error" in classified) return usage(`Error: ${classified.error}`);
-  if (classified.type !== "codex") {
-    return usage(`Error: ${name} is not a Codex account pool; cooldown clearing applies to Codex accounts only`);
+  const oauthPool = classified.type === "oauth"
+    && (name === "anthropic" || name === "google-antigravity");
+  if (classified.type !== "codex" && !oauthPool) {
+    return usage(`Error: ${name} does not support account-pool cooldown clearing`);
   }
-  const id = requestedId === "main" ? MAIN_ID : requestedId;
+  if (classified.type === "oauth" && requestedId === "main") {
+    return usage("Error: main is only valid for the openai Codex account pool");
+  }
+  const id = classified.type === "codex" && requestedId === "main" ? MAIN_ID : requestedId;
   const baseUrl = await resolveBaseUrl(deps);
   if (!baseUrl) return proxyUnreachable();
-  const response = await apiJson(deps, baseUrl, "POST", "/api/codex-auth/accounts/clear-cooldown", { id });
+  const response = classified.type === "codex"
+    ? await apiJson(deps, baseUrl, "POST", "/api/codex-auth/accounts/clear-cooldown", { id })
+    : await apiJson(deps, baseUrl, "POST", "/api/oauth/accounts/clear-cooldown", {
+      provider: name,
+      accountId: id,
+    });
   if (response.status === 0) return proxyUnreachable();
   if (response.status !== 200) return apiError(response.json, `failed to clear cooldown for ${requestedId}`);
   const cleared = response.json?.cleared === true;
diff --git a/src/cli/registry.ts b/src/cli/registry.ts
index afee044cbc..96faa73f1f 100644
--- a/src/cli/registry.ts
+++ b/src/cli/registry.ts
@@ -146,17 +146,20 @@ export const CLI_COMMANDS: CliCommandEntry[] = [
   },
   {
     name: "account",
-    usage: "ocx account <list|current|use|refresh|auto-switch|priority|login|reauth|code|cancel|remove|add-key|reset-credits|main> ...",
+    usage: "ocx account <list|current|use|refresh|auto-switch|alias|clear-cooldown|priority|login|reauth|code|cancel|remove|add-key|import|reset-credits|main> ...",
     summary: "List and switch provider accounts and API-key pools (GUI parity).",
     details: [
       "list [provider]     Codex account pool, OAuth accounts and API keys (identifiers shown masked as the API returns them).",
       "current <provider>  Show the active account or key.",
       "use <provider> <id> Switch the active credential; 'main' selects the Codex App login.",
       "refresh <provider>  Force-refresh Codex or provider quota reports.",
-      "auto-switch <provider> <on|off|status|threshold N>  Control the Codex pool threshold.",
+      "auto-switch <provider> <on|off|status|threshold N>  Control a supported account pool.",
+      "alias <provider> <id> <display-name|->  Set or clear a stored account display name.",
+      "clear-cooldown <provider> <id|main>  Clear one supported pool account's runtime cooldown; main is Codex-only.",
       "priority <provider> <id|main> [first|earlier|normal|later|last|-100..100|reset]  Selection order; omit the value to read it.",
       "remove <provider> <id> --yes  Remove a stored account or key after an existence check.",
       "add-key <provider> [--label <label>]  Add a key read only from piped stdin.",
+      "import <provider> --format <format> (--file <path>|--stdin)  Import supported account records.",
       "login/reauth/code/cancel  Run browser or manual-code auth from a headless shell.",
       "reset-credits <id|main> [--consume --yes]  Inspect or consume Codex reset credits.",
       "main <subcommand>     Manage the physical native Codex login separately from Pool routing.",
diff --git a/src/codex/pool-rotation.ts b/src/codex/pool-rotation.ts
index d0d032be06..3cc470592e 100644
--- a/src/codex/pool-rotation.ts
+++ b/src/codex/pool-rotation.ts
@@ -3,6 +3,7 @@ import type { GenerationContext } from "../lib/state-store-sweeper";
 
 export const POOL_KEY_CODEX = "codex";
 export const POOL_KEY_ANTHROPIC = "anthropic";
+export const POOL_KEY_GOOGLE_ANTIGRAVITY = "google-antigravity";
 
 interface SelectionState {
   activeKey?: string;
@@ -237,6 +238,19 @@ export function notePoolRotationFailure(poolKey: string, accountId: string): voi
   }
 }
 
+/** Remove one deleted account without disturbing the rest of the provider's rotation ring. */
+export function removePoolRotationAccount(poolKey: string, accountId: string): boolean {
+  const state = selectionState.get(poolKey);
+  if (!state) return false;
+  let removed = state.currentWeights.delete(accountId);
+  if (state.activeKey === accountId) {
+    delete state.activeKey;
+    state.successes = 0;
+    removed = true;
+  }
+  return removed;
+}
+
 /**
  * Force the next sticky/RR pick onto `accountId` (manual dashboard selection).
  * Clears sticky success counters and ring weights so the seeded account is held
@@ -260,16 +274,21 @@ export function clearPoolRotationState(poolKey?: string): void {
 export function reconcilePoolRotationState(context: GenerationContext): number {
   if (context.generation <= lastReconciledGeneration) return 0;
   const anthropicIds = new Set<string>();
+  const googleAntigravityIds = new Set<string>();
   for (const key of context.oauthAccountKeys) {
     const separator = key.indexOf("\0");
-    if (separator > 0 && key.slice(0, separator) === "anthropic") {
-      anthropicIds.add(key.slice(separator + 1));
-    }
+    if (separator <= 0) continue;
+    const provider = key.slice(0, separator);
+    const accountId = key.slice(separator + 1);
+    if (provider === "anthropic") anthropicIds.add(accountId);
+    if (provider === "google-antigravity") googleAntigravityIds.add(accountId);
   }
   let removed = 0;
   for (const [poolKey, state] of selectionState) {
     const valid = poolKey === POOL_KEY_ANTHROPIC
       ? anthropicIds
+      : poolKey === POOL_KEY_GOOGLE_ANTIGRAVITY
+        ? googleAntigravityIds
       : poolKey === POOL_KEY_CODEX || poolKey.startsWith(`${POOL_KEY_CODEX}:`)
         ? context.codexAccountIds
         : null;
diff --git a/src/config.ts b/src/config.ts
index 1308b3a64b..198d5d08d0 100644
--- a/src/config.ts
+++ b/src/config.ts
@@ -896,6 +896,13 @@ const configSchema = z.object({
   multiAgentGuidanceEnabled: z.boolean().optional(),
   // Invalid optional recovery config must not discard unrelated provider/account state.
   agentTaskRecovery: agentTaskRecoverySchema.optional().catch(undefined),
+  // Invalid hand edits disable only this opt-in pool; live writes remain strict.
+  googleAntigravityAccountPool: z.object({
+    enabled: z.boolean().optional(),
+    autoSwitchThreshold: z.number().int().min(0).max(100).optional(),
+    strategy: z.enum(["quota", "round-robin", "fill-first"]).optional(),
+    stickyLimit: z.number().int().min(1).max(100).optional(),
+  }).strict().optional().catch(undefined),
   // These selections pre-date schema validation and used to pass through as
   // unknown fields. Invalid hand edits must disable only the optional
   // delegation/native-default feature, not reject the whole config and hide
@@ -2132,6 +2139,22 @@ function oauthOpenBrowserError(value: unknown): string | null {
   return "schema_invalid: oauthOpenBrowser: must be a boolean or omitted";
 }
 
+function googleAntigravityAccountPoolError(value: unknown): string | null {
+  const raw = rawConfigRecord(value);
+  if (!raw || !Object.hasOwn(raw, "googleAntigravityAccountPool")) return null;
+  const pool = raw.googleAntigravityAccountPool;
+  if (pool === undefined || pool === null) return null;
+  if (typeof pool !== "object") return "schema_invalid: googleAntigravityAccountPool: must be an object";
+  const parse = z.object({
+    enabled: z.boolean().optional(),
+    autoSwitchThreshold: z.number().int().min(0).max(100).optional(),
+    strategy: z.enum(["quota", "round-robin", "fill-first"]).optional(),
+    stickyLimit: z.number().int().min(1).max(100).optional(),
+  }).strict().safeParse(pool);
+  if (parse.success) return null;
+  return `schema_invalid: googleAntigravityAccountPool: ${schemaDiagnosticsError(parse.error)}`;
+}
+
 /** Validate an in-memory config candidate without touching disk. Used by headless CLI import/set. */
 /**
  * Reject a loopback-listener port that collides with the proxy port (#1102).
@@ -2182,6 +2205,7 @@ export function validateConfigCandidate(value: unknown): { ok: true; config: Ocx
     ?? codexAccountPickerEnabledError(value)
     ?? emptyCompletionRetryError(value)
     ?? oauthOpenBrowserError(value)
+    ?? googleAntigravityAccountPoolError(value)
     ?? loopbackListenerPortError(value);
   if (boundaryError) return { ok: false, error: boundaryError };
   const result = configSchema.safeParse(value);
diff --git a/src/lib/state-store-registrations.ts b/src/lib/state-store-registrations.ts
index 61aad5f7b9..e1e6f13a6c 100644
--- a/src/lib/state-store-registrations.ts
+++ b/src/lib/state-store-registrations.ts
@@ -31,6 +31,10 @@ import {
   sweepExpiredXaiPermanentFailureVerdicts,
 } from "../oauth";
 import { sweepExpiredAnthropicRoutingHealth } from "../oauth/anthropic-routing";
+import {
+  reconcileGoogleAntigravityRoutingState,
+  sweepExpiredGoogleAntigravityRoutingHealth,
+} from "../oauth/google-antigravity-routing";
 import { listLiveOAuthAccountKeys, reconcileOAuthReauthState } from "../oauth/store";
 import { reconcileGuardianBackoff } from "../oauth/token-guardian";
 import { sweepExpiredApiKeyCooldowns } from "../providers/key-failover";
@@ -83,6 +87,11 @@ export const STATE_STORE_REGISTRATIONS = [
     reconcileGeneration: reconcileComboTargetCooldowns,
   },
   { name: "anthropic-routing-health", sweepExpired: sweepExpiredAnthropicRoutingHealth },
+  {
+    name: "google-antigravity-routing-health",
+    sweepExpired: sweepExpiredGoogleAntigravityRoutingHealth,
+    reconcileGeneration: reconcileGoogleAntigravityRoutingState,
+  },
   { name: "xai-refresh-verdicts", sweepExpired: sweepExpiredXaiPermanentFailureVerdicts },
   {
     name: "responses-continuation",
diff --git a/src/oauth/google-antigravity-routing.ts b/src/oauth/google-antigravity-routing.ts
new file mode 100644
index 0000000000..beb6901bdb
--- /dev/null
+++ b/src/oauth/google-antigravity-routing.ts
@@ -0,0 +1,516 @@
+/**
+ * Opt-in OAuth account routing for Google Antigravity (Cloud Code Assist).
+ *
+ * This module deliberately owns Google-specific usage families, cooldowns, and
+ * affinity. AI Studio and Vertex never import or call it.
+ */
+import { createHash } from "node:crypto";
+import { fallbackCodexAccountLogLabel } from "../codex/account-label";
+import {
+  normalizeAccountPoolStickyLimit,
+  normalizeAccountPoolStrategy,
+  clearPoolRotationState,
+  notePoolRotationFailure,
+  notePoolRotationSuccess,
+  pickRoundRobinAccount,
+  POOL_KEY_GOOGLE_ANTIGRAVITY,
+  removePoolRotationAccount,
+  seedPoolRotationAccount,
+} from "../codex/pool-rotation";
+import { retainedUtf8Bytes } from "../lib/admission";
+import { sweepExpiredOnWrite, type GenerationContext } from "../lib/state-store-sweeper";
+import { getCachedProviderAccountQuota, type ProviderQuota } from "../providers/quota";
+import type { OcxAccountPoolRotationStrategy, OcxConfig } from "../types";
+import { getAccountCredential, getAccountSet, setActiveAccount } from "./store";
+import type { OAuthAccessSnapshot } from "./index";
+
+const PROVIDER = "google-antigravity";
+const DEFAULT_AUTO_SWITCH_THRESHOLD = 80;
+const DEFAULT_COOLDOWN_MS = 60_000;
+const MAX_COOLDOWN_MS = 15 * 60_000;
+const AFFINITY_IDLE_TTL_MS = 24 * 60 * 60_000;
+const MAX_AFFINITY_ENTRIES = 2_000;
+const MAX_AFFINITY_COMPONENT_BYTES = 512;
+
+export const GOOGLE_ANTIGRAVITY_POOL_MAX_FAILOVERS_PER_REQUEST = 3;
+
+interface AccountHealth {
+  cooldownUntil: number;
+  cooldownSource: "retry-after" | "default";
+}
+
+interface AffinityEntry {
+  accountId: string;
+  lastUsedAt: number;
+}
+
+export type GoogleAntigravityAccountSelectionReason =
+  | "pool-disabled"
+  | "affinity"
+  | "active"
+  | "lowest-usage"
+  | "only-eligible"
+  | "round-robin"
+  | "fill-first"
+  | "none"
+  | "all-cooled";
+
+export interface GoogleAntigravityAccountSelection {
+  accountId: string | null;
+  reason: GoogleAntigravityAccountSelectionReason;
+}
+
+export type GoogleAntigravityQuotaRotationOutcome =
+  | { kind: "next-account"; accountId: string }
+  | { kind: "all-cooled" }
+  | { kind: "pool-disabled" }
+  | { kind: "no-eligible-account" };
+
+const upstreamHealth = new Map<string, AccountHealth>();
+const sessionAffinity = new Map<string, AffinityEntry>();
+
+function poolConfig(config: OcxConfig) {
+  const raw = config.googleAntigravityAccountPool;
+  return raw && typeof raw === "object" ? raw : {};
+}
+
+export function isGoogleAntigravityAccountPoolEnabled(config: OcxConfig): boolean {
+  return poolConfig(config).enabled === true;
+}
+
+export function googleAntigravityAutoSwitchThreshold(config: OcxConfig): number {
+  const value = poolConfig(config).autoSwitchThreshold;
+  return typeof value === "number" && Number.isInteger(value) && value >= 0 && value <= 100
+    ? value
+    : DEFAULT_AUTO_SWITCH_THRESHOLD;
+}
+
+function poolStrategy(config: OcxConfig): OcxAccountPoolRotationStrategy {
+  return normalizeAccountPoolStrategy(poolConfig(config).strategy);
+}
+
+function stickyLimit(config: OcxConfig): number {
+  return normalizeAccountPoolStickyLimit(poolConfig(config).stickyLimit);
+}
+
+function normalizeAffinityComponent(value: string | null | undefined): string {
+  const normalized = value?.trim() ?? "";
+  return normalized && retainedUtf8Bytes(normalized) <= MAX_AFFINITY_COMPONENT_BYTES ? normalized : "";
+}
+
+function pruneAffinity(now: number): void {
+  for (const [key, entry] of sessionAffinity) {
+    if (now - entry.lastUsedAt > AFFINITY_IDLE_TTL_MS) sessionAffinity.delete(key);
+  }
+  if (sessionAffinity.size <= MAX_AFFINITY_ENTRIES) return;
+  const ordered = [...sessionAffinity.entries()].sort((a, b) => a[1].lastUsedAt - b[1].lastUsedAt);
+  for (let index = 0; index < ordered.length - MAX_AFFINITY_ENTRIES; index++) {
+    sessionAffinity.delete(ordered[index]![0]);
+  }
+}
+
+function parseRetryAfterMs(value: string | null | undefined, now: number): number | undefined {
+  const text = value?.trim();
+  if (!text) return undefined;
+  if (/^\d+(?:\.\d+)?$/.test(text)) {
+    const seconds = Number(text);
+    if (Number.isFinite(seconds) && seconds > 0) {
+      return Math.min(Math.max(Math.ceil(seconds * 1000), 1), MAX_COOLDOWN_MS);
+    }
+  }
+  const timestamp = Date.parse(text);
+  const delay = timestamp - now;
+  return Number.isFinite(timestamp) && delay > 0 ? Math.min(delay, MAX_COOLDOWN_MS) : undefined;
+}
+
+export function getGoogleAntigravityAccountHealthSnapshot(
+  accountId: string,
+  now = Date.now(),
+): { cooldownUntil?: number; cooldownSource?: AccountHealth["cooldownSource"] } | null {
+  const health = upstreamHealth.get(accountId);
+  if (!health) return null;
+  if (health.cooldownUntil <= now) {
+    upstreamHealth.delete(accountId);
+    return null;
+  }
+  return { cooldownUntil: health.cooldownUntil, cooldownSource: health.cooldownSource };
+}
+
+export function clearGoogleAntigravityAccountCooldown(accountId: string): boolean {
+  return upstreamHealth.delete(accountId);
+}
+
+export function sweepExpiredGoogleAntigravityRoutingHealth(now = Date.now()): number {
+  let removed = 0;
+  for (const [accountId, health] of upstreamHealth) {
+    if (health.cooldownUntil > now) continue;
+    upstreamHealth.delete(accountId);
+    removed += 1;
+  }
+  return removed;
+}
+
+function isCooled(accountId: string, now: number): boolean {
+  return getGoogleAntigravityAccountHealthSnapshot(accountId, now) !== null;
+}
+
+function isCredentialUsable(accountId: string): boolean {
+  const credential = getAccountCredential(PROVIDER, accountId);
+  return Boolean(credential?.projectId);
+}
+
+function getUsableGoogleAntigravityAccounts(): string[] {
+  const set = getAccountSet(PROVIDER);
+  if (!set) return [];
+  return set.accounts
+    .filter(account => account.needsReauth !== true && isCredentialUsable(account.id))
+    .map(account => account.id);
+}
+
+export function getEligibleGoogleAntigravityAccounts(now = Date.now()): string[] {
+  return getUsableGoogleAntigravityAccounts().filter(accountId => !isCooled(accountId, now));
+}
+
+function modelFamily(modelId: string): "gem" | "cla" | null {
+  const normalized = modelId.toLowerCase();
+  if (normalized.includes("gemini")) return "gem";
+  if (/claude|sonnet|opus|gpt[-_]oss/.test(normalized)) return "cla";
+  return null;
+}
+
+function familyWindowMatches(family: "gem" | "cla", label: string): boolean {
+  const normalized = label.trim().toLowerCase();
+  return family === "gem"
+    ? /(^|\b)gem(?:ini)?(\b|$)/.test(normalized)
+    : /(^|\b)cla(?:ude)?(\b|$)|sonnet|opus|gpt[-_ ]?oss/.test(normalized);
+}
+
+/** Maximum finite used percent relevant to this CCA model family; null means unknown. */
+export function googleAntigravityUsageScore(modelId: string, quota: ProviderQuota | null | undefined): number | null {
+  if (!quota) return null;
+  const canonical = [quota.fiveHourPercent, quota.weeklyPercent, quota.monthlyPercent]
+    .filter((value): value is number => typeof value === "number" && Number.isFinite(value));
+  const family = modelFamily(modelId);
+  const custom = family
+    ? (quota.customWindows ?? [])
+      .filter(window => familyWindowMatches(family, window.label))
+      .map(window => window.percent)
+      .filter((value): value is number => typeof value === "number" && Number.isFinite(value))
+    : [];
+  const relevant = [...canonical, ...custom];
+  if (relevant.length === 0) return null;
+  return Math.max(0, Math.min(100, Math.max(...relevant)));
+}
+
+function usageScore(accountId: string, modelId: string): number | null {
+  return googleAntigravityUsageScore(modelId, getCachedProviderAccountQuota(PROVIDER, accountId));
+}
+
+function activeUsable(accountId: string, now: number): boolean {
+  const set = getAccountSet(PROVIDER);
+  return Boolean(set?.accounts.some(account => account.id === accountId && account.needsReauth !== true))
+    && !isCooled(accountId, now)
+    && isCredentialUsable(accountId);
+}
+
+function pickLowestUsage(modelId: string, excludeId: string | undefined, now: number): string | null {
+  const eligible = getEligibleGoogleAntigravityAccounts(now).filter(id => id !== excludeId);
+  if (eligible.length === 0) return null;
+  const known = eligible
+    .map(id => ({ id, score: usageScore(id, modelId) }))
+    .filter((row): row is { id: string; score: number } => row.score !== null);
+  if (known.length === 0) return eligible[0] ?? null;
+  known.sort((a, b) => a.score - b.score);
+  return known[0]!.id;
+}
+
+function underThreshold(config: OcxConfig, accountId: string, modelId: string): boolean {
+  const threshold = googleAntigravityAutoSwitchThreshold(config);
+  if (threshold <= 0) return true;
+  const score = usageScore(accountId, modelId);
+  return score === null || score < threshold;
+}
+
+function nextFillFirst(config: OcxConfig, modelId: string, afterId: string, eligible: string[]): string | null {
+  const set = getAccountSet(PROVIDER);
+  const all = [...(set?.accounts.map(account => account.id) ?? eligible)].sort((a, b) => a.localeCompare(b));
+  const start = all.indexOf(afterId);
+  let fallback: string | null = null;
+  for (let step = 1; step <= all.length; step++) {
+    const candidate = all[(Math.max(start, -1) + step) % all.length]!;
+    if (!eligible.includes(candidate)) continue;
+    fallback ??= candidate;
+    if (underThreshold(config, candidate, modelId)) return candidate;
+  }
+  return fallback;
+}
+
+function pickStrategyAccount(
+  config: OcxConfig,
+  modelId: string,
+  now: number,
+): { accountId: string; reason: "round-robin" | "fill-first" } | null {
+  const strategy = poolStrategy(config);
+  const eligible = getEligibleGoogleAntigravityAccounts(now);
+  if (strategy === "round-robin") {
+    const limit = stickyLimit(config);
+    const accountId = pickRoundRobinAccount(POOL_KEY_GOOGLE_ANTIGRAVITY, eligible, limit);
+    if (!accountId) return null;
+    notePoolRotationSuccess(POOL_KEY_GOOGLE_ANTIGRAVITY, accountId, limit);
+    return { accountId, reason: "round-robin" };
+  }
+  if (strategy === "fill-first") {
+    const active = getAccountSet(PROVIDER)?.activeAccountId;
+    const accountId = active && eligible.includes(active) && underThreshold(config, active, modelId)
+      ? active
+      : nextFillFirst(config, modelId, active ?? "", eligible);
+    return accountId ? { accountId, reason: "fill-first" } : null;
+  }
+  return null;
+}
+
+export function resolveGoogleAntigravityAccountForSession(
+  sessionKey: string | null | undefined,
+  modelId: string,
+  config: OcxConfig,
+  now = Date.now(),
+): GoogleAntigravityAccountSelection {
+  pruneAffinity(now);
+  const set = getAccountSet(PROVIDER);
+  if (!set || set.accounts.length === 0) return { accountId: null, reason: "none" };
+  if (!isGoogleAntigravityAccountPoolEnabled(config)) {
+    return { accountId: set.activeAccountId, reason: "pool-disabled" };
+  }
+
+  const key = normalizeAffinityComponent(sessionKey);
+  const affined = key ? sessionAffinity.get(key) : undefined;
+  if (affined) {
+    if (now - affined.lastUsedAt <= AFFINITY_IDLE_TTL_MS && activeUsable(affined.accountId, now)) {
+      return { accountId: affined.accountId, reason: "affinity" };
+    }
+    sessionAffinity.delete(key);
+  }
+
+  const active = set.activeAccountId;
+  const activeOk = activeUsable(active, now);
+  const strategy = poolStrategy(config);
+  if (!key && (strategy === "round-robin" || strategy === "fill-first") && activeOk) {
+    return { accountId: active, reason: "active" };
+  }
+
+  const strategyPick = pickStrategyAccount(config, modelId, now);
+  if (strategyPick) {
+    return strategyPick;
+  }
+
+  const threshold = googleAntigravityAutoSwitchThreshold(config);
+  let accountId: string | null = null;
+  let reason: GoogleAntigravityAccountSelectionReason = "none";
+  const activeScore = usageScore(active, modelId);
+  if (threshold > 0 && activeOk && (activeScore === null || activeScore < threshold)) {
+    accountId = active;
+    reason = "active";
+  } else if (threshold > 0) {
+    accountId = pickLowestUsage(modelId, undefined, now);
+    if (accountId) reason = accountId === active ? "active" : "lowest-usage";
+  } else if (activeOk) {
+    accountId = active;
+    reason = "active";
+  } else {
+    accountId = pickLowestUsage(modelId, active, now);
+    if (accountId) reason = "only-eligible";
+  }
+
+  if (!accountId) {
+    return {
+      accountId: null,
+      reason: set.accounts.some(account => isCooled(account.id, now)) ? "all-cooled" : "none",
+    };
+  }
+  return { accountId, reason };
+}
+
+export function bindGoogleAntigravitySessionAffinity(
+  sessionKey: string | null | undefined,
+  accountId: string,
+  now = Date.now(),
+): void {
+  const key = normalizeAffinityComponent(sessionKey);
+  if (!key || !normalizeAffinityComponent(accountId)) return;
+  sessionAffinity.set(key, { accountId, lastUsedAt: now });
+  pruneAffinity(now);
+}
+
+function clearAffinityForAccount(accountId: string): void {
+  for (const [key, entry] of sessionAffinity) {
+    if (entry.accountId === accountId) sessionAffinity.delete(key);
+  }
+}
+
+/** Drop every provider-owned runtime reference to a deleted Google OAuth account. */
+export function clearGoogleAntigravityRoutingStateForAccount(accountId: string): void {
+  upstreamHealth.delete(accountId);
+  clearAffinityForAccount(accountId);
+  removePoolRotationAccount(POOL_KEY_GOOGLE_ANTIGRAVITY, accountId);
+}
+
+export function reconcileGoogleAntigravityRoutingState(context: GenerationContext): number {
+  const validAccountIds = new Set<string>();
+  for (const key of context.oauthAccountKeys) {
+    const separator = key.indexOf("\0");
+    if (separator > 0 && key.slice(0, separator) === PROVIDER) {
+      validAccountIds.add(key.slice(separator + 1));
+    }
+  }
+  let removed = 0;
+  for (const accountId of upstreamHealth.keys()) {
+    if (validAccountIds.has(accountId)) continue;
+    upstreamHealth.delete(accountId);
+    removed += 1;
+  }
+  for (const [key, entry] of sessionAffinity) {
+    if (validAccountIds.has(entry.accountId)) continue;
+    sessionAffinity.delete(key);
+    removed += 1;
+  }
+  return removed;
+}
+
+function pickAlternate(
+  config: OcxConfig,
+  modelId: string,
+  failedAccountId: string,
+  now: number,
+): string | null {
+  const eligible = getEligibleGoogleAntigravityAccounts(now).filter(id => id !== failedAccountId);
+  const strategy = poolStrategy(config);
+  if (strategy === "round-robin") {
+    return pickRoundRobinAccount(POOL_KEY_GOOGLE_ANTIGRAVITY, eligible, stickyLimit(config));
+  }
+  if (strategy === "fill-first") return nextFillFirst(config, modelId, failedAccountId, eligible);
+  return pickLowestUsage(modelId, failedAccountId, now);
+}
+
+/** Cool the failed account and propose an explicit outcome; the caller commits affinity only when it dispatches. */
+export function rotateGoogleAntigravityAccountOnQuotaError(
+  config: OcxConfig,
+  failedAccountId: string,
+  retryAfterHeader: string | null | undefined,
+  modelId: string,
+  _sessionKey?: string | null,
+  now = Date.now(),
+): GoogleAntigravityQuotaRotationOutcome {
+  if (!coolGoogleAntigravityAccountOnQuotaError(config, failedAccountId, retryAfterHeader, now)) {
+    return { kind: "pool-disabled" };
+  }
+  const next = pickAlternate(config, modelId, failedAccountId, now);
+  if (!next) {
+    const usable = getUsableGoogleAntigravityAccounts();
+    if (usable.length > 0 && usable.every(accountId => isCooled(accountId, now))) {
+      console.warn("[google-antigravity-pool] all eligible OAuth accounts are in cooldown; returning quota response");
+      return { kind: "all-cooled" };
+    }
+    return { kind: "no-eligible-account" };
+  }
+  return { kind: "next-account", accountId: next };
+}
+
+/** Classify one surfaced quota response even when the request cannot dispatch another account. */
+export function coolGoogleAntigravityAccountOnQuotaError(
+  config: OcxConfig,
+  failedAccountId: string,
+  retryAfterHeader: string | null | undefined,
+  now = Date.now(),
+): boolean {
+  if (!isGoogleAntigravityAccountPoolEnabled(config)) return false;
+  const retryAfterMs = parseRetryAfterMs(retryAfterHeader, now);
+  upstreamHealth.set(failedAccountId, {
+    cooldownUntil: now + (retryAfterMs ?? DEFAULT_COOLDOWN_MS),
+    cooldownSource: retryAfterMs === undefined ? "default" : "retry-after",
+  });
+  clearAffinityForAccount(failedAccountId);
+  notePoolRotationFailure(POOL_KEY_GOOGLE_ANTIGRAVITY, failedAccountId);
+  sweepExpiredOnWrite(now);
+  return true;
+}
+
+export function getGoogleAntigravityPoolRetryAfterSeconds(now = Date.now()): number | null {
+  const set = getAccountSet(PROVIDER);
+  let earliest: number | null = null;
+  for (const account of set?.accounts ?? []) {
+    const until = getGoogleAntigravityAccountHealthSnapshot(account.id, now)?.cooldownUntil;
+    if (until !== undefined && (earliest === null || until < earliest)) earliest = until;
+  }
+  return earliest !== null && earliest > now ? Math.max(1, Math.ceil((earliest - now) / 1000)) : null;
+}
+
+export function googleAntigravityAllCooledError(now = Date.now()): {
+  status: 429;
+  type: "rate_limit_error";
+  message: string;
+  retryAfter?: string;
+} {
+  const retryAfterSeconds = getGoogleAntigravityPoolRetryAfterSeconds(now);
+  return {
+    status: 429,
+    type: "rate_limit_error",
+    message: "All Google Antigravity OAuth accounts are temporarily rate-limited",
+    ...(retryAfterSeconds !== null ? { retryAfter: String(retryAfterSeconds) } : {}),
+  };
+}
+
+export function promoteGoogleAntigravityActiveAccount(accountId: string): void {
+  void setActiveAccount(PROVIDER, accountId).catch(() => { /* best effort */ });
+}
+
+/** Resolve the selected account's bearer and CCA project as one generation-bound snapshot. */
+export async function getGoogleAntigravityPoolAccessSnapshot(
+  accountId: string,
+): Promise<OAuthAccessSnapshot & { projectId: string }> {
+  const { getValidAccessSnapshotForAccount } = await import("./index");
+  const snapshot = await getValidAccessSnapshotForAccount(PROVIDER, accountId);
+  if (!snapshot.projectId) {
+    throw new Error("selected Google Antigravity account is missing its Cloud Code Assist project id");
+  }
+  return { ...snapshot, projectId: snapshot.projectId };
+}
+
+export function resetGoogleAntigravityRoutingForManualSelection(accountId: string): void {
+  sessionAffinity.clear();
+  seedPoolRotationAccount(POOL_KEY_GOOGLE_ANTIGRAVITY, accountId);
+}
+
+export function formatGoogleAntigravityAccountOrdinal(accountId: string): string {
+  return fallbackCodexAccountLogLabel(accountId);
+}
+
+export function formatGoogleAntigravityProviderForLog(
+  providerName: string,
+  accountId: string | null | undefined,
+): string {
+  return accountId ? `${providerName}-${formatGoogleAntigravityAccountOrdinal(accountId)}` : providerName;
+}
+
+export function googleAntigravitySessionKeyFromParts(input: {
+  sessionIdHeader?: string | null;
+  threadIdHeader?: string | null;
+  promptCacheKey?: string | null;
+  clientThreadId?: string | null;
+  promptCacheKeyIsSharedCohort?: boolean;
+}): string | null {
+  const preferred = (input.clientThreadId ?? input.sessionIdHeader ?? input.threadIdHeader ?? "").trim();
+  if (preferred) return preferred.length <= 128 ? preferred : createHash("sha256").update(preferred).digest("hex");
+  if (input.promptCacheKeyIsSharedCohort) return null;
+  const cacheKey = input.promptCacheKey?.trim() ?? "";
+  return !cacheKey ? null : cacheKey.length <= 128 ? cacheKey : createHash("sha256").update(cacheKey).digest("hex");
+}
+
+export function clearGoogleAntigravityAccountPoolState(): void {
+  upstreamHealth.clear();
+  sessionAffinity.clear();
+  clearPoolRotationState(POOL_KEY_GOOGLE_ANTIGRAVITY);
+}
+
+export function googleAntigravitySessionAffinitySizeForTests(): number {
+  return sessionAffinity.size;
+}
diff --git a/src/providers/quota.ts b/src/providers/quota.ts
index 63c74b9cd6..25705813f7 100644
--- a/src/providers/quota.ts
+++ b/src/providers/quota.ts
@@ -10,7 +10,12 @@ import { isMainAccountIdentityGenerationLive } from "../codex/main-account-cache
 import { MAIN_CODEX_ACCOUNT_ID } from "../codex/main-account";
 import { codexPlanKey } from "../codex/plan";
 import { resolveEnvValue } from "../config";
-import { getValidAccessToken, getValidAccessTokenForAccount } from "../oauth";
+import {
+  getValidAccessToken,
+  getValidAccessTokenForAccount,
+  getValidAccessSnapshotForAccount,
+  getValidAccessTokenSnapshot,
+} from "../oauth";
 import { getAccountCredential, getAccountSet, getCredential } from "../oauth/store";
 import { antigravityUserAgent } from "../adapters/client-fingerprint";
 import { apiKeyPoolEntryId } from "./api-keys";
@@ -1452,7 +1457,7 @@ export interface ProviderAccountQuota {
 
 /** Providers whose per-account quota can be probed. Extend as other OAuth APIs are covered. */
 export function supportsPerAccountQuota(provider: string): boolean {
-  return provider === "anthropic";
+  return provider === "anthropic" || provider === "google-antigravity";
 }
 
 function accountCacheKey(provider: string, accountId: string): string {
@@ -1571,8 +1576,9 @@ async function fetchAccountQuota(
 
   const probe = (async (): Promise<AccountQuotaCacheEntry> => {
     try {
-      const token = await getTokenForAccountQuotaProbe(provider, accountId);
-      const quota = await fetchAnthropicUsageQuota(token);
+      const quota = provider === "google-antigravity"
+        ? await fetchAntigravityAccountQuota(accountId)
+        : await fetchAnthropicUsageQuota(await getTokenForAccountQuotaProbe(provider, accountId));
       if (!quota) {
         // Preserve last-good bars and mark unavailable; advance TTL so failures
         // negative-cache instead of re-probing on every GUI poll.
@@ -2129,29 +2135,8 @@ function antigravityUsedPercent(quotaInfo: Record<string, unknown>): number | un
   return normalizePercent(100 - remaining);
 }
 
-async function fetchAntigravityQuota(provider: string, config: OcxProviderConfig): Promise<ProviderQuotaReport | null> {
-  const credential = getCredential("google-antigravity");
-  if (!credential?.projectId) return null;
-  let accessToken: string;
-  try {
-    accessToken = await getValidAccessToken("google-antigravity");
-  } catch {
-    return null;
-  }
-  const baseUrl = (config.baseUrl || "https://daily-cloudcode-pa.googleapis.com").replace(/\/+$/, "");
-  const response = await fetch(`${baseUrl}/v1internal:fetchAvailableModels`, {
-    method: "POST",
-    headers: {
-      Accept: "application/json",
-      "Content-Type": "application/json",
-      "User-Agent": antigravityUserAgent(),
-      Authorization: `Bearer ${accessToken}`,
-    },
-    body: JSON.stringify({ project: credential.projectId }),
-    signal: AbortSignal.timeout(REQUEST_TIMEOUT_MS),
-  });
-  if (!response.ok) return null;
-  const body = asRecord(await readQuotaJson(response));
+function parseAntigravityQuotaPayload(value: unknown): ProviderQuota | null {
+  const body = asRecord(value);
   const models = asRecord(body?.models);
   if (!models) return null;
 
@@ -2161,13 +2146,18 @@ async function fetchAntigravityQuota(provider: string, config: OcxProviderConfig
     if (!modelInfo) continue;
     for (const quotaInfo of quotaInfoEntries(modelInfo)) {
       const label = classifyAntigravityFamily(modelId, modelInfo, quotaInfo);
-      if (!label || windows.has(label)) continue;
+      if (!label) continue;
       const percent = antigravityUsedPercent(quotaInfo);
       if (percent === undefined) continue;
+      const resetAt = normalizeResetAt(quotaInfo.resetTime);
+      const current = windows.get(label);
+      // Multiple model rows may share a family. Keep the maximum used percent
+      // so routing never treats a constrained family as healthier than it is.
+      if (current && current.percent >= percent) continue;
       windows.set(label, {
         label,
         percent,
-        ...(normalizeResetAt(quotaInfo.resetTime) !== undefined ? { resetAt: normalizeResetAt(quotaInfo.resetTime) } : {}),
+        ...(resetAt !== undefined ? { resetAt } : {}),
       });
     }
   }
@@ -2176,11 +2166,52 @@ async function fetchAntigravityQuota(provider: string, config: OcxProviderConfig
     const window = windows.get(label);
     return window ? [window] : [];
   });
-  if (customWindows.length === 0) return null;
-  return report(provider, "google-antigravity:fetchAvailableModels", {
-    customWindows,
-    updatedAt: Date.now(),
+  return customWindows.length > 0 ? { customWindows, updatedAt: Date.now() } : null;
+}
+
+async function fetchAntigravityAccountQuota(accountId: string): Promise<ProviderQuota | null> {
+  const snapshot = await getValidAccessSnapshotForAccount("google-antigravity", accountId);
+  if (!snapshot.projectId) return null;
+  const baseUrl = getProviderRegistryEntry("google-antigravity")?.baseUrl;
+  if (!baseUrl) return null;
+  const response = await fetch(`${baseUrl.replace(/\/+$/, "")}/v1internal:fetchAvailableModels`, {
+    method: "POST",
+    headers: {
+      Accept: "application/json",
+      "Content-Type": "application/json",
+      "User-Agent": antigravityUserAgent(),
+      Authorization: `Bearer ${snapshot.accessToken}`,
+    },
+    body: JSON.stringify({ project: snapshot.projectId }),
+    signal: AbortSignal.timeout(REQUEST_TIMEOUT_MS),
+  });
+  if (!response.ok) return null;
+  return parseAntigravityQuotaPayload(await readQuotaJson(response));
+}
+
+async function fetchAntigravityQuota(provider: string, config: OcxProviderConfig): Promise<ProviderQuotaReport | null> {
+  let snapshot: Awaited<ReturnType<typeof getValidAccessTokenSnapshot>>;
+  try {
+    snapshot = await getValidAccessTokenSnapshot("google-antigravity");
+  } catch {
+    return null;
+  }
+  if (!snapshot.projectId) return null;
+  const baseUrl = (config.baseUrl || "https://daily-cloudcode-pa.googleapis.com").replace(/\/+$/, "");
+  const response = await fetch(`${baseUrl}/v1internal:fetchAvailableModels`, {
+    method: "POST",
+    headers: {
+      Accept: "application/json",
+      "Content-Type": "application/json",
+      "User-Agent": antigravityUserAgent(),
+      Authorization: `Bearer ${snapshot.accessToken}`,
+    },
+    body: JSON.stringify({ project: snapshot.projectId }),
+    signal: AbortSignal.timeout(REQUEST_TIMEOUT_MS),
   });
+  if (!response.ok) return null;
+  const quota = parseAntigravityQuotaPayload(await readQuotaJson(response));
+  return quota ? report(provider, "google-antigravity:fetchAvailableModels", quota) : null;
 }
 
 async function maybeFetchProviderQuota(
diff --git a/src/routing/analytics.ts b/src/routing/analytics.ts
index 6dc217e734..76cdca98e3 100644
--- a/src/routing/analytics.ts
+++ b/src/routing/analytics.ts
@@ -120,6 +120,7 @@ const COOLDOWN_RECOVERY_KINDS = new Set([
   "oauth-401",
   "anthropic-oauth-429",
   "oauth-account-429",
+  "google-antigravity-oauth-quota",
 ]);
 
 function percentile(sorted: number[], p: number): number | undefined {
diff --git a/src/server/management/oauth-account-routes.ts b/src/server/management/oauth-account-routes.ts
index ee57515b41..4aad499a90 100644
--- a/src/server/management/oauth-account-routes.ts
+++ b/src/server/management/oauth-account-routes.ts
@@ -95,6 +95,24 @@ async function readJsonBody(req: Request): Promise<Record<string, unknown> | nul
   }
 }
 
+type OAuthPoolProvider = "anthropic" | "google-antigravity";
+type OAuthPoolConfig = NonNullable<OcxConfig["anthropicAccountPool"]>;
+
+function oauthPoolProvider(value: string): value is OAuthPoolProvider {
+  return value === "anthropic" || value === "google-antigravity";
+}
+
+function oauthPoolConfig(config: OcxConfig, provider: OAuthPoolProvider): OAuthPoolConfig {
+  return provider === "anthropic"
+    ? config.anthropicAccountPool ?? {}
+    : config.googleAntigravityAccountPool ?? {};
+}
+
+function setOAuthPoolConfig(config: OcxConfig, provider: OAuthPoolProvider, pool: OAuthPoolConfig): void {
+  if (provider === "anthropic") config.anthropicAccountPool = pool;
+  else config.googleAntigravityAccountPool = pool;
+}
+
 /**
  * The single place key-name rules live. The config read schema is deliberately
  * permissive so an existing config can never become unloadable; this is the write
@@ -167,15 +185,10 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
           reconcileLiveStateStores();
         },
       });
-      // Open the browser server-side (the proxy runs on the user's machine) — the GUI's
-      // window.open is popup-blocked because it runs after an await, not a direct click.
-      //
-      // The operator can decline, which is the only way to finish a login in a
-      // browser profile other than the OS default, or on a different machine
-      // than the proxy. Declining changes nothing else: the URL is still
-      // returned below and every login surface renders it with a copy button.
       const { shouldOpenBrowserForLogin } = await import("../../oauth/open-browser-choice");
       if (authUrl && !deviceCode && shouldOpenBrowserForLogin(body.openBrowser, config)) {
+        // Open the browser server-side (the proxy runs on the user's machine) — the GUI's
+        // window.open is popup-blocked because it runs after an await, not a direct click.
         const { openUrl } = await import("../../lib/open-url");
         openUrl(authUrl);
       }
@@ -305,6 +318,9 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
     if (provider === "anthropic") {
       const { resetAnthropicRoutingForManualSelection } = await import("../../oauth/anthropic-routing");
       resetAnthropicRoutingForManualSelection(body.accountId);
+    } else if (provider === "google-antigravity") {
+      const { resetGoogleAntigravityRoutingForManualSelection } = await import("../../oauth/google-antigravity-routing");
+      resetGoogleAntigravityRoutingForManualSelection(body.accountId);
     }
     const { clearModelCache } = await import("../../codex/model-cache");
     const { clearGatherRoutedModelsInflight } = await import("../../codex/catalog");
@@ -315,11 +331,13 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
     return jsonResponse({ ok: true, provider, activeAccountId: body.accountId });
   }
 
-  // Opt-in Anthropic OAuth account pool (#294): enable/threshold/strategy + clear cooldown.
+  // Opt-in OAuth account pools: enable/threshold/strategy + clear cooldown.
   if (url.pathname === "/api/oauth/accounts/pool" && req.method === "GET") {
     const provider = (url.searchParams.get("provider") ?? "").trim().toLowerCase();
-    if (provider !== "anthropic") return jsonResponse({ error: "pool config is only supported for anthropic" }, 400);
-    const pool = config.anthropicAccountPool ?? {};
+    if (!oauthPoolProvider(provider)) {
+      return jsonResponse({ error: "pool config is only supported for anthropic and google-antigravity" }, 400);
+    }
+    const pool = oauthPoolConfig(config, provider);
     return jsonResponse({
       provider,
       enabled: pool.enabled === true,
@@ -342,13 +360,22 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
       stickyLimit?: unknown;
     };
     const provider = typeof body.provider === "string" ? body.provider.trim().toLowerCase() : "";
-    if (provider !== "anthropic") return jsonResponse({ error: "pool config is only supported for anthropic" }, 400);
-    let enabled = config.anthropicAccountPool?.enabled === true;
+    if (!oauthPoolProvider(provider)) {
+      return jsonResponse({ error: "pool config is only supported for anthropic and google-antigravity" }, 400);
+    }
+    const currentPool = oauthPoolConfig(config, provider);
+    let enabled = currentPool.enabled === true;
     if (body.enabled !== undefined) {
       if (typeof body.enabled !== "boolean") return jsonResponse({ error: "enabled must be a boolean" }, 400);
       enabled = body.enabled;
     }
-    let threshold = config.anthropicAccountPool?.autoSwitchThreshold ?? 80;
+    const currentThreshold = currentPool.autoSwitchThreshold;
+    let threshold = typeof currentThreshold === "number"
+      && Number.isInteger(currentThreshold)
+      && currentThreshold >= 0
+      && currentThreshold <= 100
+      ? currentThreshold
+      : 80;
     if (body.autoSwitchThreshold !== undefined) {
       if (
         typeof body.autoSwitchThreshold !== "number"
@@ -360,7 +387,9 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
       }
       threshold = body.autoSwitchThreshold;
     }
-    let strategy = config.anthropicAccountPool?.strategy;
+    let strategy = currentPool.strategy === undefined
+      ? undefined
+      : parseAccountPoolStrategy(currentPool.strategy) ?? "quota";
     if (body.strategy !== undefined) {
       const parsed = parseAccountPoolStrategy(body.strategy);
       if (parsed === null) {
@@ -368,7 +397,9 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
       }
       strategy = parsed;
     }
-    let stickyLimit = config.anthropicAccountPool?.stickyLimit;
+    let stickyLimit = currentPool.stickyLimit === undefined
+      ? undefined
+      : parseAccountPoolStickyLimit(currentPool.stickyLimit) ?? 1;
     if (body.stickyLimit !== undefined) {
       const parsed = parseAccountPoolStickyLimit(body.stickyLimit);
       if (parsed === null) {
@@ -376,12 +407,12 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
       }
       stickyLimit = parsed;
     }
-    config.anthropicAccountPool = {
+    setOAuthPoolConfig(config, provider, {
       enabled,
       autoSwitchThreshold: threshold,
       ...(strategy !== undefined ? { strategy } : {}),
       ...(stickyLimit !== undefined ? { stickyLimit } : {}),
-    };
+    });
     saveConfigPreservingClaudeCode(config);
     reconcileLiveStateStores();
     return jsonResponse({
@@ -398,10 +429,17 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
     const body = await readManagementJsonBodyOr(req, {}) as { provider?: unknown; accountId?: unknown };
     const provider = typeof body.provider === "string" ? body.provider.trim().toLowerCase() : "";
     const accountId = typeof body.accountId === "string" ? body.accountId.trim() : "";
-    if (provider !== "anthropic") return jsonResponse({ error: "clear-cooldown is only supported for anthropic" }, 400);
+    if (!oauthPoolProvider(provider)) {
+      return jsonResponse({ error: "clear-cooldown is only supported for anthropic and google-antigravity" }, 400);
+    }
     if (!accountId) return jsonResponse({ error: "missing accountId" }, 400);
-    const { clearAnthropicAccountCooldown } = await import("../../oauth/anthropic-routing");
-    const cleared = clearAnthropicAccountCooldown(accountId);
+    const { getAccountSet } = await import("../../oauth/store");
+    if (!getAccountSet(provider)?.accounts.some(account => account.id === accountId)) {
+      return jsonResponse({ error: "account not found" }, 400);
+    }
+    const cleared = provider === "anthropic"
+      ? (await import("../../oauth/anthropic-routing")).clearAnthropicAccountCooldown(accountId)
+      : (await import("../../oauth/google-antigravity-routing")).clearGoogleAntigravityAccountCooldown(accountId);
     return jsonResponse({ ok: true, cleared });
   }
 
@@ -480,6 +518,9 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
       const { clearAnthropicAccountCooldown, clearAnthropicSessionAffinityForAccount } = await import("../../oauth/anthropic-routing");
       clearAnthropicAccountCooldown(id);
       clearAnthropicSessionAffinityForAccount(id);
+    } else if (provider === "google-antigravity") {
+      const { clearGoogleAntigravityRoutingStateForAccount } = await import("../../oauth/google-antigravity-routing");
+      clearGoogleAntigravityRoutingStateForAccount(id);
     }
     if (!getAccountSet(provider)) clearLoginState(provider);
     const { clearModelCache } = await import("../../codex/model-cache");
diff --git a/src/server/responses/core.ts b/src/server/responses/core.ts
index 08a679894d..d69aba5435 100644
--- a/src/server/responses/core.ts
+++ b/src/server/responses/core.ts
@@ -121,6 +121,18 @@ import {
   isGenericOAuthFailoverEnabled,
   rotateGenericOAuthAccountOn429,
 } from "../../oauth/generic-account-failover";
+import {
+  bindGoogleAntigravitySessionAffinity,
+  formatGoogleAntigravityProviderForLog,
+  getGoogleAntigravityPoolAccessSnapshot,
+  googleAntigravityAllCooledError,
+  GOOGLE_ANTIGRAVITY_POOL_MAX_FAILOVERS_PER_REQUEST,
+  googleAntigravitySessionKeyFromParts,
+  isGoogleAntigravityAccountPoolEnabled,
+  promoteGoogleAntigravityActiveAccount,
+  resolveGoogleAntigravityAccountForSession,
+  rotateGoogleAntigravityAccountOnQuotaError,
+} from "../../oauth/google-antigravity-routing";
 import { buildWebSearchTool, planWebSearch, runWithWebSearch, shouldResolveOpenAiWebSearchSidecar } from "../../web-search";
 import { buildImageTool, buildVideoTool, planImageBridge, planVideoBridge, runWithImageBridge, clampImageMaxRounds, IMAGE_GEN_TOOL_NAME, VIDEO_GEN_TOOL_NAME } from "../../images";
 import { describeImagesInPlace, isModelTextOnly, planVisionSidecar, resolveOpenAiVisionModel, shouldResolveOpenAiVisionSidecar, stripImagesInPlace } from "../../vision";
@@ -327,14 +339,14 @@ import {
 import {
   collectDeclaredNamelessClientCallTypes,
   collectDeclaredWireToolNames,
-  collectProviderExecutedCallTypes,
   createUndeclaredToolCallGuardBlockRewrite,
+  collectProviderExecutedCallTypes,
+  type ProviderExecutedCallType,
   currentTurnWireToolCatalogBody,
   hasExplicitWireToolCatalog,
   undeclaredToolCallMessage,
   undeclaredToolCallName,
   undeclaredToolCallNameInResponse,
-  type ProviderExecutedCallType,
 } from "../responses-undeclared-tool-guard";
 import { createGithubCopilotResponsesBlockRewrite } from "../github-copilot-responses-repair";
 import { responsesJsonToSseStream } from "../responses-json-events";
@@ -826,6 +838,32 @@ export function upstreamHostCircuitOpenResponse(retryAfterSeconds: number): Resp
   );
 }
 
+function googleAntigravityAllCooledResponse(): Response {
+  const error = googleAntigravityAllCooledError();
+  return formatErrorResponse(
+    error.status,
+    error.type,
+    error.message,
+    error.retryAfter !== undefined ? { retryAfter: error.retryAfter } : undefined,
+  );
+}
+
+function assertNeverGoogleAntigravityRotation(value: never): never {
+  throw new Error("Unhandled Google Antigravity account rotation outcome");
+}
+
+const GOOGLE_ANTIGRAVITY_FAILOVER_LIMIT_MESSAGE =
+  "Google Antigravity quota exhausted after bounded account failover";
+
+function googleAntigravityFailoverLimitResponse(status: 429 | 402, retryAfter: string | null): Response {
+  return formatErrorResponse(
+    status,
+    status === 429 ? "rate_limit_error" : "insufficient_quota",
+    GOOGLE_ANTIGRAVITY_FAILOVER_LIMIT_MESSAGE,
+    retryAfter !== null ? { retryAfter } : undefined,
+  );
+}
+
 function normalizeCodexUnsupportedModelDetail(value: string): string {
   return value.trim().replace(/\s+/gu, " ").toLocaleLowerCase("en-US");
 }
@@ -2851,6 +2889,21 @@ async function handleResponsesInner(
       promptCacheKeyIsSharedCohort: options.promptCacheKeyIsSharedCohort === true,
     })
     : null;
+  let googleAntigravityPoolAccountId: string | null = null;
+  let googleAntigravityPoolFailovers = 0;
+  const googleAntigravityPoolActive = route.providerName === "google-antigravity"
+    && route.provider.authMode === "oauth"
+    && route.provider.googleMode === "cloud-code-assist"
+    && isGoogleAntigravityAccountPoolEnabled(config);
+  const googleAntigravitySessionKey = googleAntigravityPoolActive
+    ? googleAntigravitySessionKeyFromParts({
+      sessionIdHeader: sessionIdHeaderFromRequest(req.headers),
+      threadIdHeader: req.headers.get("thread-id"),
+      promptCacheKey: typeof parsed.options.promptCacheKey === "string" ? parsed.options.promptCacheKey : null,
+      clientThreadId: typeof parsed._clientThreadId === "string" ? parsed._clientThreadId : null,
+      promptCacheKeyIsSharedCohort: options.promptCacheKeyIsSharedCohort === true,
+    })
+    : null;
   if (route.provider.authMode === "oauth") {
     try {
       if (route.providerName === "anthropic" && isAnthropicAccountPoolEnabled(config)) {
@@ -2873,6 +2926,36 @@ async function handleResponsesInner(
         promoteAnthropicActiveAccount(selection.accountId);
         route.provider = { ...route.provider, apiKey: accessToken };
         logCtx.provider = formatAnthropicProviderForLog("anthropic", selection.accountId, config);
+      } else if (googleAntigravityPoolActive) {
+        const selection = resolveGoogleAntigravityAccountForSession(
+          googleAntigravitySessionKey,
+          route.modelId,
+          config,
+        );
+        if (!selection.accountId) {
+          if (selection.reason === "all-cooled") {
+            return googleAntigravityAllCooledResponse();
+          }
+          return formatErrorResponse(
+            401,
+            "authentication_error",
+            "No eligible Google Antigravity OAuth account available",
+          );
+        }
+        const snapshot = await getGoogleAntigravityPoolAccessSnapshot(selection.accountId);
+        googleAntigravityPoolAccountId = selection.accountId;
+        replayOAuthCredentialSnapshot = {
+          accountId: snapshot.accountId,
+          generation: snapshot.generation,
+        };
+        bindGoogleAntigravitySessionAffinity(googleAntigravitySessionKey, selection.accountId);
+        promoteGoogleAntigravityActiveAccount(selection.accountId);
+        route.provider = {
+          ...route.provider,
+          apiKey: snapshot.accessToken,
+          project: snapshot.projectId,
+        };
+        logCtx.provider = formatGoogleAntigravityProviderForLog("google-antigravity", selection.accountId);
       } else {
         const resolved = await getValidAccessTokenSnapshot(route.providerName);
         replayOAuthCredentialSnapshot = {
@@ -5179,6 +5262,80 @@ async function handleResponsesInner(
           break;
         }
       }
+      // Opt-in Google Antigravity OAuth pool: only the exact Cloud Code Assist
+      // provider owns this branch. A final 429 (after adapter-local retries) or
+      // surfaced 402 cools the failed account and rebuilds token + project from
+      // one generation-bound snapshot before dispatching the next account.
+      googleAntigravityFailover: while (
+        (upstreamResponse.status === 429 || upstreamResponse.status === 402)
+        && googleAntigravityPoolAccountId
+        && googleAntigravityPoolActive
+      ) {
+        const rotation = rotateGoogleAntigravityAccountOnQuotaError(
+          config,
+          googleAntigravityPoolAccountId,
+          upstreamResponse.headers.get("retry-after"),
+          route.modelId,
+          googleAntigravitySessionKey,
+        );
+        let nextAccountId: string;
+        switch (rotation.kind) {
+          case "all-cooled":
+            try { void upstreamResponse.body?.cancel().catch(() => {}); } catch { /* already consumed/closed */ }
+            return googleAntigravityAllCooledResponse();
+          case "pool-disabled":
+          case "no-eligible-account":
+            break googleAntigravityFailover;
+          case "next-account":
+            if (googleAntigravityPoolFailovers >= GOOGLE_ANTIGRAVITY_POOL_MAX_FAILOVERS_PER_REQUEST) {
+              try { void upstreamResponse.body?.cancel().catch(() => {}); } catch { /* already consumed/closed */ }
+              return googleAntigravityFailoverLimitResponse(
+                upstreamResponse.status === 402 ? 402 : 429,
+                upstreamResponse.headers.get("retry-after"),
+              );
+            }
+            nextAccountId = rotation.accountId;
+            break;
+          default:
+            return assertNeverGoogleAntigravityRotation(rotation);
+        }
+        try { void upstreamResponse.body?.cancel().catch(() => {}); } catch { /* already consumed/closed */ }
+        try {
+          const snapshot = await getGoogleAntigravityPoolAccessSnapshot(nextAccountId);
+          bindGoogleAntigravitySessionAffinity(googleAntigravitySessionKey, nextAccountId);
+          googleAntigravityPoolAccountId = nextAccountId;
+          googleAntigravityPoolFailovers += 1;
+          replayOAuthCredentialSnapshot = {
+            accountId: snapshot.accountId,
+            generation: snapshot.generation,
+          };
+          route.provider = {
+            ...route.provider,
+            apiKey: snapshot.accessToken,
+            project: snapshot.projectId,
+          };
+          invalidateSameTargetRequest();
+          promoteGoogleAntigravityActiveAccount(nextAccountId);
+          logCtx.provider = formatGoogleAntigravityProviderForLog("google-antigravity", nextAccountId);
+          activeAdapter = resolveAdapter(
+            resolveWireProtocolOverride(route.providerName, route.modelId, route.provider, inboundWire),
+            config.cacheRetention,
+          );
+          bindRouteReasoningReplayScope({
+            parsed,
+            providerName: route.providerName,
+            provider: route.provider,
+            adapterName: activeAdapter.name,
+            oauthCredentialSnapshot: replayOAuthCredentialSnapshot,
+          });
+          sealRequestAttemptIdentity(logCtx.activeAttempt, logCtx.provider, activeAdapter.name, logCtx.accountLogLabel);
+          const result = await rebuildAndRefetch("google-antigravity-oauth-quota");
+          if ("failed" in result) return result.failed;
+          upstreamResponse = result;
+        } catch {
+          break;
+        }
+      }
       // Generic OAuth account failover (#2568) for providers with no pool of their own. Opt-in
       // and a strict no-op otherwise, so a single-account install and every existing config are
       // unchanged. Codex and Anthropic are excluded by isGenericFailoverProvider — their pools
@@ -5340,6 +5497,10 @@ async function handleResponsesInner(
   const terminalGuardEnabled = (activeAdapter.name === "anthropic"
       || (activeAdapter.name === "openai-chat" && route.provider.terminalContinuationGuard === true))
     && !options.comboAttempt && !routedCompaction;
+  let googleAntigravityContinuationAllCooled = false;
+  const googleAntigravityContinuationFailoverLimit = {
+    value: null as { status: 429 | 402; retryAfter: string | null } | null,
+  };
   /**
    * One bounded internal re-ask for Anthropic end_turn-without-tool-call turns. Replays the
    * continuation on a 429 with the same-key retry budget (hoisted per request), then falls
@@ -5565,6 +5726,99 @@ async function handleResponsesInner(
           }
         }
       }
+      if (
+        (response.status === 429 || response.status === 402)
+        && googleAntigravityPoolAccountId
+        && googleAntigravityPoolActive
+      ) {
+        const rotation = rotateGoogleAntigravityAccountOnQuotaError(
+          config,
+          googleAntigravityPoolAccountId,
+          response.headers.get("retry-after"),
+          route.modelId,
+          googleAntigravitySessionKey,
+        );
+        switch (rotation.kind) {
+          case "all-cooled": {
+            try { void response.body?.cancel().catch(() => {}); } catch { /* already consumed/closed */ }
+            googleAntigravityContinuationAllCooled = true;
+            const error = googleAntigravityAllCooledError();
+            yield {
+              type: "error",
+              status: error.status,
+              errorType: error.type,
+              message: error.message,
+            };
+            return;
+          }
+          case "pool-disabled":
+          case "no-eligible-account":
+            break;
+          case "next-account": {
+            if (googleAntigravityPoolFailovers >= GOOGLE_ANTIGRAVITY_POOL_MAX_FAILOVERS_PER_REQUEST) {
+              try { void response.body?.cancel().catch(() => {}); } catch { /* already consumed/closed */ }
+              const status = response.status as 429 | 402;
+              googleAntigravityContinuationFailoverLimit.value = {
+                status,
+                retryAfter: response.headers.get("retry-after"),
+              };
+              yield {
+                type: "error",
+                status,
+                errorType: status === 429 ? "rate_limit_error" : "insufficient_quota",
+                message: GOOGLE_ANTIGRAVITY_FAILOVER_LIMIT_MESSAGE,
+              };
+              return;
+            }
+            const nextAccountId = rotation.accountId;
+            try { void response.body?.cancel().catch(() => {}); } catch { /* already closed */ }
+            try {
+              const snapshot = await getGoogleAntigravityPoolAccessSnapshot(nextAccountId);
+              bindGoogleAntigravitySessionAffinity(googleAntigravitySessionKey, nextAccountId);
+              googleAntigravityPoolAccountId = nextAccountId;
+              googleAntigravityPoolFailovers += 1;
+              replayOAuthCredentialSnapshot = {
+                accountId: snapshot.accountId,
+                generation: snapshot.generation,
+              };
+              route.provider = {
+                ...route.provider,
+                apiKey: snapshot.accessToken,
+                project: snapshot.projectId,
+              };
+              invalidateSameTargetRequest();
+              promoteGoogleAntigravityActiveAccount(nextAccountId);
+              logCtx.provider = formatGoogleAntigravityProviderForLog("google-antigravity", nextAccountId);
+              activeAdapter = resolveAdapter(
+                resolveWireProtocolOverride(route.providerName, route.modelId, route.provider, inboundWire),
+                config.cacheRetention,
+              );
+              bindRouteReasoningReplayScope({
+                parsed: nextParsed,
+                providerName: route.providerName,
+                provider: route.provider,
+                adapterName: activeAdapter.name,
+                oauthCredentialSnapshot: replayOAuthCredentialSnapshot,
+              });
+              bindRouteReasoningReplayScope({
+                parsed,
+                providerName: route.providerName,
+                provider: route.provider,
+                adapterName: activeAdapter.name,
+                oauthCredentialSnapshot: replayOAuthCredentialSnapshot,
+              });
+              sealRequestAttemptIdentity(logCtx.activeAttempt, logCtx.provider, activeAdapter.name, logCtx.accountLogLabel);
+              nextContinuationRecoveryKind = "google-antigravity-oauth-quota";
+              continue;
+            } catch {
+              // Fall through to the bounded continuation error surface.
+            }
+            break;
+          }
+          default:
+            return assertNeverGoogleAntigravityRotation(rotation);
+        }
+      }
       if (shouldAttemptImageTierRetry({
         status: response.status,
         adapterName: activeAdapter.name,
@@ -5744,6 +5998,16 @@ async function handleResponsesInner(
     } finally {
       cleanupUpstreamAbort();
     }
+    if (googleAntigravityContinuationAllCooled) {
+      return googleAntigravityAllCooledResponse();
+    }
+    const continuationFailoverLimit = googleAntigravityContinuationFailoverLimit.value;
+    if (continuationFailoverLimit) {
+      return googleAntigravityFailoverLimitResponse(
+        continuationFailoverLimit.status,
+        continuationFailoverLimit.retryAfter,
+      );
+    }
     const { toolNsMap, declaredToolNames, toolParameterSchemas, freeformToolNames, toolSearchToolNames } = toolBridgeMaps;
     let providerState: OcxProviderContinuationState | undefined;
     const json = buildResponseJSON(events, parsed._responseModelId ?? parsed.modelId, {
diff --git a/src/types/config.ts b/src/types/config.ts
index 1262b68d7c..2b9a4d1acd 100644
--- a/src/types/config.ts
+++ b/src/types/config.ts
@@ -620,6 +620,16 @@ export interface OcxConfig {
     /** Successful new-session binds retained on one round-robin selection. Default 1; range 1..100. */
     stickyLimit?: number;
   };
+  /** Opt-in quota auto-switching and cooldown pooling for Google Antigravity (Cloud Code Assist). */
+  googleAntigravityAccountPool?: {
+    enabled?: boolean;
+    /** Usage % threshold for new-session auto-pick. Default 80. 0 = disabled (affinity/active only). */
+    autoSwitchThreshold?: number;
+    /** New-session rotation strategy. Default quota (today's behaviour). */
+    strategy?: OcxAccountPoolRotationStrategy;
+    /** Successful new-session binds retained on one round-robin selection. Default 1; range 1..100. */
+    stickyLimit?: number;
+  };
   /**
    * Generic OAuth multi-account 429 failover (#2568). Presence-driven by default.
    *
diff --git a/src/usage/log.ts b/src/usage/log.ts
index 3406ba9ba1..b48e189895 100644
--- a/src/usage/log.ts
+++ b/src/usage/log.ts
@@ -29,6 +29,7 @@ export type AttemptRecoveryKind =
   | "rate-limit-429"
   | "anthropic-oauth-429"
   | "oauth-account-429"
+  | "google-antigravity-oauth-quota"
   | "image-413"
   | "opaque-blob-rejection"
   | "empty-completion";
@@ -221,6 +222,7 @@ const ATTEMPT_RECOVERY_KINDS = new Set<AttemptRecoveryKind>([
   "rate-limit-429",
   "anthropic-oauth-429",
   "oauth-account-429",
+  "google-antigravity-oauth-quota",
   "image-413",
   "opaque-blob-rejection",
   "empty-completion",
diff --git a/tests/account-pool-management-api.test.ts b/tests/account-pool-management-api.test.ts
index 4dbe399269..012ccada4d 100644
--- a/tests/account-pool-management-api.test.ts
+++ b/tests/account-pool-management-api.test.ts
@@ -6,8 +6,19 @@ import { join } from "node:path";
 import { handleCodexAuthAPI } from "../src/codex/auth-api";
 import { saveConfig } from "../src/config";
 import { startServer } from "../src/server";
+import { handleOauthAccountRoutes } from "../src/server/management/oauth-account-routes";
 import type { OcxConfig } from "../src/types";
 import { installIsolatedCodexHome, type IsolatedCodexHome } from "./helpers/isolated-codex-home";
+import {
+  bindGoogleAntigravitySessionAffinity,
+  getEligibleGoogleAntigravityAccounts,
+  getGoogleAntigravityAccountHealthSnapshot,
+  googleAntigravitySessionAffinitySizeForTests,
+  resetGoogleAntigravityRoutingForManualSelection,
+  resolveGoogleAntigravityAccountForSession,
+  rotateGoogleAntigravityAccountOnQuotaError,
+} from "../src/oauth/google-antigravity-routing";
+import { getAccountSet, removeAccount, saveCredential } from "../src/oauth/store";
 
 function makeCodexConfig(overrides: Partial<OcxConfig> = {}): OcxConfig {
   return {
@@ -150,6 +161,12 @@ describe("Anthropic account pool strategy management API", () => {
       defaultProvider: "anthropic",
       providers: {
         anthropic: { adapter: "anthropic", baseUrl: "https://api.anthropic.com", authMode: "oauth" },
+        "google-antigravity": {
+          adapter: "google",
+          baseUrl: "https://daily-cloudcode-pa.googleapis.com",
+          authMode: "oauth",
+          googleMode: "cloud-code-assist",
+        },
       },
     } as OcxConfig;
   }
@@ -167,6 +184,13 @@ describe("Anthropic account pool strategy management API", () => {
           { id: "aaaa1111", credential: { access: "t1", refresh: "r1", expires: 9999999999999, email: "a@example.com", accountId: "acct-1" } },
         ],
       },
+      "google-antigravity": {
+        activeAccountId: "google-a",
+        accounts: [
+          { id: "google-a", credential: { access: "ga", refresh: "gra", expires: 9999999999999, projectId: "project-a" } },
+          { id: "google-b", credential: { access: "gb", refresh: "grb", expires: 9999999999999, projectId: "project-b" } },
+        ],
+      },
     }), { mode: 0o600 });
   });
 
@@ -352,4 +376,227 @@ describe("Anthropic account pool strategy management API", () => {
       await server.stop(true);
     }
   });
+
+  test("Google Antigravity GET uses backward-compatible pool defaults", async () => {
+    const server = startServer(0);
+    try {
+      const response = await fetch(new URL("/api/oauth/accounts/pool?provider=google-antigravity", server.url));
+      expect(response.status).toBe(200);
+      expect(await response.json()).toMatchObject({
+        provider: "google-antigravity",
+        enabled: false,
+        autoSwitchThreshold: 80,
+        strategy: "quota",
+        stickyLimit: 1,
+      });
+    } finally {
+      await server.stop(true);
+    }
+  });
+
+  test("Google Antigravity PUT validates and persists generic pool settings", async () => {
+    const server = startServer(0);
+    try {
+      const invalid = await fetch(new URL("/api/oauth/accounts/pool", server.url), {
+        method: "PUT",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({
+          provider: "google-antigravity",
+          enabled: "yes",
+          autoSwitchThreshold: 101,
+          strategy: "weighted",
+        }),
+      });
+      expect(invalid.status).toBe(400);
+
+      const put = await fetch(new URL("/api/oauth/accounts/pool", server.url), {
+        method: "PUT",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({
+          provider: "google-antigravity",
+          enabled: true,
+          autoSwitchThreshold: 65,
+          strategy: "fill-first",
+          stickyLimit: 6,
+        }),
+      });
+      expect(put.status).toBe(200);
+      expect(await put.json()).toMatchObject({
+        ok: true,
+        provider: "google-antigravity",
+        enabled: true,
+        autoSwitchThreshold: 65,
+        strategy: "fill-first",
+        stickyLimit: 6,
+      });
+
+      const get = await fetch(new URL("/api/oauth/accounts/pool?provider=google-antigravity", server.url));
+      expect(await get.json()).toMatchObject({ enabled: true, strategy: "fill-first", stickyLimit: 6 });
+    } finally {
+      await server.stop(true);
+    }
+  });
+
+  test("Google Antigravity partial updates normalize invalid current pool fields before saving", async () => {
+    const invalidConfig = baseConfig() as OcxConfig & {
+      googleAntigravityAccountPool: Record<string, unknown>;
+    };
+    invalidConfig.googleAntigravityAccountPool = {
+      enabled: false,
+      autoSwitchThreshold: "broken",
+      strategy: "weighted",
+      stickyLimit: 0,
+    };
+    const req = new Request("http://localhost/api/oauth/accounts/pool", {
+      method: "PATCH",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ provider: "google-antigravity", enabled: true }),
+    });
+    const response = await handleOauthAccountRoutes({
+      req,
+      url: new URL(req.url),
+      config: invalidConfig,
+      deps: {},
+      convergeCodexCatalog: async () => ({ status: "unchanged" }),
+      syncClaudeAgentDefsBestEffort: async () => {},
+    });
+
+    expect(response?.status).toBe(200);
+    expect(invalidConfig.googleAntigravityAccountPool).toEqual({
+      enabled: true,
+      autoSwitchThreshold: 80,
+      strategy: "quota",
+      stickyLimit: 1,
+    });
+  });
+
+  test("OAuth clear-cooldown rejects unknown accounts but stays idempotent for stored accounts", async () => {
+    const server = startServer(0);
+    try {
+      for (const { provider, knownId } of [
+        { provider: "google-antigravity", knownId: "google-b" },
+        { provider: "anthropic", knownId: "aaaa1111" },
+      ]) {
+        const unknownId = `unknown-${provider}-account`;
+        const unknown = await fetch(new URL("/api/oauth/accounts/clear-cooldown", server.url), {
+          method: "POST",
+          headers: { "Content-Type": "application/json" },
+          body: JSON.stringify({ provider, accountId: unknownId }),
+        });
+        expect(unknown.status).toBe(400);
+        const unknownBody = await unknown.json() as { error?: string };
+        expect(unknownBody).toMatchObject({ error: "account not found" });
+        expect(JSON.stringify(unknownBody)).not.toContain(unknownId);
+
+        const known = await fetch(new URL("/api/oauth/accounts/clear-cooldown", server.url), {
+          method: "POST",
+          headers: { "Content-Type": "application/json" },
+          body: JSON.stringify({ provider, accountId: knownId }),
+        });
+        expect(known.status).toBe(200);
+        expect(await known.json()).toEqual({ ok: true, cleared: false });
+      }
+    } finally {
+      await server.stop(true);
+    }
+  });
+
+  test("Google clear-cooldown and manual active selection use provider-owned routing state", async () => {
+    const server = startServer(0);
+    try {
+      const config = baseConfig();
+      config.googleAntigravityAccountPool = { enabled: true };
+      const set = getAccountSet("google-antigravity")!;
+      const aId = set.accounts.find(account => account.id === "google-a")!.id;
+      const bId = set.accounts.find(account => account.id === "google-b")!.id;
+      bindGoogleAntigravitySessionAffinity("manual-thread", aId);
+      expect(rotateGoogleAntigravityAccountOnQuotaError(
+        config, aId, "60", "gemini-3.7-flash", "other-thread",
+      )).toEqual({ kind: "next-account", accountId: bId });
+      expect(getGoogleAntigravityAccountHealthSnapshot(aId)).not.toBeNull();
+
+      const clear = await fetch(new URL("/api/oauth/accounts/clear-cooldown", server.url), {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({ provider: "google-antigravity", accountId: aId }),
+      });
+      expect(clear.status).toBe(200);
+      expect(await clear.json()).toMatchObject({ ok: true, cleared: true });
+      expect(getGoogleAntigravityAccountHealthSnapshot(aId)).toBeNull();
+
+      const select = await fetch(new URL("/api/oauth/accounts/active", server.url), {
+        method: "PUT",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({ provider: "google-antigravity", accountId: bId }),
+      });
+      expect(select.status).toBe(200);
+      expect(resolveGoogleAntigravityAccountForSession(
+        "manual-thread", "gemini-3.7-flash", config,
+      )).toMatchObject({ accountId: bId });
+    } finally {
+      await server.stop(true);
+    }
+  });
+
+  test("deleting Google account state makes a deterministic re-add immediately eligible", async () => {
+    const config = baseConfig();
+    config.googleAntigravityAccountPool = { enabled: true, strategy: "round-robin", stickyLimit: 5 };
+    saveConfig(config);
+    await removeAccount("google-antigravity", "google-a");
+    await removeAccount("google-antigravity", "google-b");
+    await saveCredential("google-antigravity", {
+      access: "delete-access",
+      refresh: "delete-refresh",
+      expires: Date.now() + 3_600_000,
+      accountId: "delete-identity",
+      email: "delete@example.test",
+      projectId: "delete-project",
+    });
+    const accountId = getAccountSet("google-antigravity")!.accounts
+      .find(account => account.credential.accountId === "delete-identity")!.id;
+    rotateGoogleAntigravityAccountOnQuotaError(config, accountId, "120", "gemini-3.7-flash");
+    resetGoogleAntigravityRoutingForManualSelection(accountId);
+    bindGoogleAntigravitySessionAffinity("deleted-account-thread", accountId);
+    expect(getGoogleAntigravityAccountHealthSnapshot(accountId)).not.toBeNull();
+    expect(googleAntigravitySessionAffinitySizeForTests()).toBe(1);
+
+    const server = startServer(0);
+    try {
+      const removed = await fetch(new URL(
+        `/api/oauth/accounts?provider=google-antigravity&id=${encodeURIComponent(accountId)}`,
+        server.url,
+      ), { method: "DELETE" });
+      expect(removed.status).toBe(200);
+      expect(getGoogleAntigravityAccountHealthSnapshot(accountId)).toBeNull();
+      expect(googleAntigravitySessionAffinitySizeForTests()).toBe(0);
+
+      await saveCredential("google-antigravity", {
+        access: "replacement-access",
+        refresh: "replacement-refresh",
+        expires: Date.now() + 3_600_000,
+        accountId: "replacement-identity",
+        email: "replacement@example.test",
+        projectId: "replacement-project",
+      });
+      const replacementId = getAccountSet("google-antigravity")!.accounts
+        .find(account => account.credential.accountId === "replacement-identity")!.id;
+      await saveCredential("google-antigravity", {
+        access: "delete-access-new",
+        refresh: "delete-refresh-new",
+        expires: Date.now() + 3_600_000,
+        accountId: "delete-identity",
+        email: "delete@example.test",
+        projectId: "delete-project",
+      });
+      const readdedId = getAccountSet("google-antigravity")!.accounts
+        .find(account => account.credential.accountId === "delete-identity")!.id;
+      expect(readdedId).toBe(accountId);
+      expect(getEligibleGoogleAntigravityAccounts()).toContain(accountId);
+      expect(resolveGoogleAntigravityAccountForSession(
+        "post-delete-rotation", "gemini-3.7-flash", config,
+      )).toMatchObject({ accountId: replacementId, reason: "round-robin" });
+    } finally {
+      await server.stop(true);
+    }
+  });
 });
diff --git a/tests/cli-account.test.ts b/tests/cli-account.test.ts
index 8a0110d679..c3b4fcb931 100644
--- a/tests/cli-account.test.ts
+++ b/tests/cli-account.test.ts
@@ -48,6 +48,10 @@ let server: ReturnType<typeof Bun.serve>;
 let baseUrl = "";
 let activeCodexAccountId: string | null = "chatgpt_1";
 let autoSwitchThreshold = 80;
+const oauthPoolSettings: Record<string, { enabled: boolean; autoSwitchThreshold: number; strategy: string; stickyLimit: number }> = {
+  anthropic: { enabled: false, autoSwitchThreshold: 80, strategy: "quota", stickyLimit: 1 },
+  "google-antigravity": { enabled: false, autoSwitchThreshold: 80, strategy: "quota", stickyLimit: 1 },
+};
 let activeReadFailure: { status: number; error: string } | null = null;
 let oauthListFailure: { provider: string; status: number; error: string } | null = null;
 let keyListFailure: { provider: string; status: number; error: string } | null = null;
@@ -88,6 +92,12 @@ function fixtureConfig(): OcxConfig {
         baseUrl: "https://api.anthropic.com",
         authMode: "oauth",
       },
+      "google-antigravity": {
+        adapter: "google",
+        baseUrl: "https://daily-cloudcode-pa.googleapis.com",
+        authMode: "oauth",
+        googleMode: "cloud-code-assist",
+      },
       kiro: {
         adapter: "anthropic",
         baseUrl: "https://q.us-east-1.amazonaws.com",
@@ -128,7 +138,7 @@ function json(body: unknown, status = 200): Response {
 
 async function mockManagementApi(req: Request): Promise<Response> {
   const url = new URL(req.url);
-  const body = req.method === "PUT" || req.method === "POST" ? await req.json() : undefined;
+  const body = req.method === "PUT" || req.method === "PATCH" || req.method === "POST" ? await req.json() : undefined;
   requests.push({ method: req.method, path: url.pathname, search: url.search, body });
 
   if (req.method === "GET" && url.pathname === "/api/codex-auth/accounts") {
@@ -170,6 +180,10 @@ async function mockManagementApi(req: Request): Promise<Response> {
     return json({ ok: true, id: payload.id, priority: account.priority });
   }
 
+  if (req.method === "POST" && url.pathname === "/api/codex-auth/accounts/clear-cooldown") {
+    return json({ ok: true, cleared: false });
+  }
+
   if (url.pathname === "/api/codex-auth/active") {
     if (req.method === "PUT") {
       const accountId = (body as { accountId?: string }).accountId;
@@ -191,7 +205,27 @@ async function mockManagementApi(req: Request): Promise<Response> {
   }
 
   if (req.method === "GET" && url.pathname === "/api/oauth/providers") {
-    return json({ providers: ["anthropic", "kiro", "xai"] });
+    return json({ providers: ["anthropic", "google-antigravity", "kiro", "xai"] });
+  }
+
+  if (url.pathname === "/api/oauth/accounts/pool") {
+    const provider = req.method === "GET"
+      ? url.searchParams.get("provider") ?? ""
+      : String((body as { provider?: unknown } | undefined)?.provider ?? "");
+    const settings = oauthPoolSettings[provider];
+    if (!settings) return json({ error: "unsupported pool provider" }, 400);
+    if (req.method === "GET") return json({ provider, ...settings });
+    if (req.method === "PUT" || req.method === "PATCH") {
+      const update = body as { enabled?: boolean; autoSwitchThreshold?: number; strategy?: string; stickyLimit?: number };
+      Object.assign(settings, update);
+      return json({ ok: true, provider, ...settings });
+    }
+  }
+
+  if (req.method === "POST" && url.pathname === "/api/oauth/accounts/clear-cooldown") {
+    const accountId = String((body as { accountId?: unknown } | undefined)?.accountId ?? "");
+    if (accountId === "unknown-google-account") return json({ error: "account not found" }, 400);
+    return json({ ok: true, cleared: true });
   }
 
   if (req.method === "GET" && url.pathname === "/api/provider-quotas") {
@@ -407,6 +441,8 @@ afterAll(() => {
 beforeEach(() => {
   activeCodexAccountId = "chatgpt_1";
   autoSwitchThreshold = 80;
+  oauthPoolSettings.anthropic = { enabled: false, autoSwitchThreshold: 80, strategy: "quota", stickyLimit: 1 };
+  oauthPoolSettings["google-antigravity"] = { enabled: false, autoSwitchThreshold: 80, strategy: "quota", stickyLimit: 1 };
   activeReadFailure = null;
   oauthListFailure = null;
   keyListFailure = null;
@@ -779,19 +815,91 @@ describe("ocx account CLI (issue #180 matrix)", () => {
     });
   });
 
-  test("21: auto-switch rejects wrong providers, invalid thresholds and missing providers", async () => {
-    const wrongProvider = await run(["auto-switch", "anthropic", "on"]);
+  test("21: auto-switch rejects unsupported providers, invalid thresholds and missing providers", async () => {
+    const wrongProvider = await run(["auto-switch", "kiro", "on"]);
     const invalidThreshold = await run(["auto-switch", "openai", "threshold", "101"]);
     const missingProvider = await run(["auto-switch"]);
 
     expect(wrongProvider.code).toBe(1);
-    expect(wrongProvider.stderr).toContain("auto-switch only applies to the openai Codex account pool");
+    expect(wrongProvider.stderr).toContain("does not support account-pool auto-switch");
     expect(invalidThreshold.code).toBe(1);
     expect(invalidThreshold.stderr).toContain("integer 0-100");
     expect(missingProvider.code).toBe(1);
     expect(missingProvider.stderr).toContain("Usage:");
   });
 
+  test("OAuth auto-switch uses the generic pool API for Anthropic and Google Antigravity", async () => {
+    const googleOn = await run(["auto-switch", "google-antigravity", "on"]);
+    const googleThreshold = await run(["auto-switch", "google-antigravity", "threshold", "61"]);
+    const googleStatus = await run(["auto-switch", "google-antigravity", "status", "--json"]);
+    const anthropicOff = await run(["auto-switch", "anthropic", "off"]);
+
+    expect(googleOn.code).toBe(0);
+    expect(googleThreshold.code).toBe(0);
+    expect(anthropicOff.code).toBe(0);
+    expect(JSON.parse(googleStatus.stdout)).toMatchObject({
+      provider: "google-antigravity",
+      enabled: true,
+      autoSwitchThreshold: 61,
+    });
+    expect(requests.filter(request => request.path === "/api/oauth/accounts/pool").map(request => ({
+      method: request.method,
+      search: request.search,
+      body: request.body,
+    }))).toEqual([
+      { method: "PUT", search: "", body: { provider: "google-antigravity", enabled: true } },
+      { method: "PUT", search: "", body: { provider: "google-antigravity", enabled: true, autoSwitchThreshold: 61 } },
+      { method: "GET", search: "?provider=google-antigravity", body: undefined },
+      { method: "PUT", search: "", body: { provider: "anthropic", enabled: false } },
+    ]);
+  });
+
+  test("clear-cooldown uses the generic OAuth route for Google and keeps priority unsupported", async () => {
+    const clear = await run(["clear-cooldown", "google-antigravity", "google-account-1", "--json"]);
+    const priority = await run(["priority", "google-antigravity", "google-account-1", "first"]);
+
+    expect(clear.code).toBe(0);
+    expect(JSON.parse(clear.stdout)).toEqual({
+      ok: true,
+      provider: "google-antigravity",
+      id: "google-account-1",
+      cleared: true,
+    });
+    expect(requests.find(request => request.path === "/api/oauth/accounts/clear-cooldown")?.body).toEqual({
+      provider: "google-antigravity",
+      accountId: "google-account-1",
+    });
+    expect(priority.code).toBe(1);
+    expect(priority.stderr).toContain("only applies to the openai Codex account pool");
+  });
+
+  test("Google clear-cooldown returns nonzero for an unknown account without echoing its id", async () => {
+    const unknownId = "unknown-google-account";
+    const result = await run(["clear-cooldown", "google-antigravity", unknownId, "--json"]);
+
+    expect(result.code).toBe(1);
+    expect(result.stderr).toContain("account not found");
+    expect(result.output).not.toContain(unknownId);
+  });
+
+  test("Google clear-cooldown rejects the Codex-only main alias without exposing its sentinel", async () => {
+    const result = await run(["clear-cooldown", "google-antigravity", "main", "--json"]);
+
+    expect(result.code).toBe(1);
+    expect(result.stderr).toContain("main");
+    expect(result.output).not.toContain("__main__");
+    expect(requests.some(request => request.path === "/api/oauth/accounts/clear-cooldown")).toBe(false);
+  });
+
+  test("OpenAI clear-cooldown keeps main as the Codex-only account alias", async () => {
+    const result = await run(["clear-cooldown", "openai", "main"]);
+
+    expect(result.code).toBe(0);
+    expect(result.stdout).toContain("no active cooldown for main");
+    expect(requests.find(request => request.path === "/api/codex-auth/accounts/clear-cooldown")?.body)
+      .toEqual({ id: "__main__" });
+  });
+
   test("22: remove without --yes prints the re-run hint and sends no request", async () => {
     // Recording fetchImpl proves no HTTP call is even attempted — the --yes
     // guard fires at arg-parse time, before resolveBaseUrl (Carver C-gate).
@@ -988,11 +1096,29 @@ describe("ocx account CLI (issue #180 matrix)", () => {
     logs.length = 0;
     printSubcommandUsage("account");
     const help = logs.join("\n");
-    for (const command of ["refresh", "auto-switch", "remove", "add-key"]) {
+    for (const command of ["refresh", "auto-switch", "clear-cooldown", "remove", "add-key"]) {
       expect(help).toContain(command);
     }
   });
 
+  test("account command and registry help expose the same shipped subcommand inventory", async () => {
+    const command = await run([]);
+    expect(command.code).toBe(1);
+
+    logs.length = 0;
+    printSubcommandUsage("account");
+    const registry = logs.join("\n");
+    const inventory = [
+      "list", "current", "use", "refresh", "auto-switch", "alias", "priority", "remove",
+      "clear-cooldown", "add-key", "import", "login", "reauth", "code", "cancel",
+      "reset-credits", "main",
+    ];
+    for (const subcommand of inventory) {
+      expect(command.stderr, `command usage must include ${subcommand}`).toContain(subcommand);
+      expect(registry, `registry help must include ${subcommand}`).toContain(subcommand);
+    }
+  });
+
   test("C-gate fold: add-key redacts a key containing JSON-escaped characters", async () => {
     const key = 'sk-"x\\test';
     const human = await run(
diff --git a/tests/config.test.ts b/tests/config.test.ts
index 6ebae31319..c2626f87bc 100644
--- a/tests/config.test.ts
+++ b/tests/config.test.ts
@@ -151,6 +151,46 @@ describe("opencodex config defaults", () => {
     });
   });
 
+  test("Google Antigravity pool config is strict, validates, and round-trips safely", () => {
+    const defaults = getDefaultConfig();
+    const pool = {
+      enabled: true,
+      autoSwitchThreshold: 65,
+      strategy: "round-robin" as const,
+      stickyLimit: 4,
+    };
+    expect(validateConfigCandidate({ ...defaults, googleAntigravityAccountPool: pool })).toMatchObject({
+      ok: true,
+      config: { googleAntigravityAccountPool: pool },
+    });
+    for (const invalid of [
+      { enabled: "yes" },
+      { autoSwitchThreshold: 101 },
+      { autoSwitchThreshold: 1.5 },
+      { strategy: "weighted" },
+      { stickyLimit: 0 },
+      { stickyLimit: 101 },
+      { future: true },
+    ]) {
+      const result = validateConfigCandidate({ ...defaults, googleAntigravityAccountPool: invalid });
+      expect(result.ok).toBe(false);
+      if (!result.ok) expect(result.error).toContain("googleAntigravityAccountPool");
+    }
+
+    saveConfig({ ...defaults, googleAntigravityAccountPool: pool });
+    expect(loadConfig().googleAntigravityAccountPool).toEqual(pool);
+
+    writeConfig({
+      ...defaults,
+      providers: { custom: { adapter: "openai-chat", baseUrl: "https://example.test/v1" } },
+      defaultProvider: "custom",
+      googleAntigravityAccountPool: { enabled: "yes", autoSwitchThreshold: 999 },
+    });
+    const loaded = loadConfig();
+    expect(loaded.providers.custom).toBeDefined();
+    expect(loaded.googleAntigravityAccountPool).toBeUndefined();
+  });
+
   test("usage and MCP config overrides change the effective bound while defaults remain compatible", () => {
     const defaults = getDefaultConfig();
     expect(defaults.managementUsageMaxReadBytes).toBe(64 * 1024 * 1024);
@@ -1652,40 +1692,6 @@ describe("opencodex config defaults", () => {
     expect(readConfigDiagnostics().error).toContain("providers.custom.modelMaxInputTokens");
   });
 
-  test("disk config validates per-model auto-compaction budgets with native exact ids", () => {
-    writeConfig({
-      port: 10100,
-      providers: {
-        custom: {
-          adapter: "openai-chat",
-          baseUrl: "https://example.test/v1",
-          modelAutoCompactTokenLimits: { model: 1.5 },
-        },
-      },
-      defaultProvider: "custom",
-    });
-    expect(readConfigDiagnostics().source).toBe("fallback");
-    expect(readConfigDiagnostics().error).toContain("providers.custom.modelAutoCompactTokenLimits");
-
-    rmSync(testDir, { recursive: true, force: true });
-    mkdirSync(testDir, { recursive: true });
-    writeConfig({
-      port: 10100,
-      providers: {
-        openai: {
-          adapter: "openai-responses",
-          baseUrl: "https://chatgpt.com/backend-api/codex",
-          authMode: "forward",
-          codexAccountMode: "direct",
-          modelAutoCompactTokenLimits: { "team/gpt-5.6-sol": 64_000 },
-        },
-      },
-      defaultProvider: "openai",
-    });
-    expect(readConfigDiagnostics().source).toBe("fallback");
-    expect(readConfigDiagnostics().error).toContain("exact supported native model id");
-  });
-
   test("disk config preserves valid OpenRouter routing and rejects invalid destinations", () => {
     writeConfig({
       port: 10100,
diff --git a/tests/google-antigravity-account-pool-request.test.ts b/tests/google-antigravity-account-pool-request.test.ts
new file mode 100644
index 0000000000..a41a622a45
--- /dev/null
+++ b/tests/google-antigravity-account-pool-request.test.ts
@@ -0,0 +1,537 @@
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
+import { mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { saveConfig } from "../src/config";
+import {
+  clearGoogleAntigravityAccountPoolState,
+  formatGoogleAntigravityProviderForLog,
+  getGoogleAntigravityAccountHealthSnapshot,
+  googleAntigravitySessionAffinitySizeForTests,
+} from "../src/oauth/google-antigravity-routing";
+import { getAccountSet, markAccountNeedsReauth, saveCredential, setActiveAccount } from "../src/oauth/store";
+import { clearAccountQuotaCache } from "../src/providers/quota";
+import { clearRequestLogsForTests, getRequestLogEntries } from "../src/server/request-log";
+import { startServer } from "../src/server";
+import type { OcxConfig } from "../src/types";
+import { installIsolatedCodexHome, type IsolatedCodexHome } from "./helpers/isolated-codex-home";
+
+interface Dispatch {
+  authorization: string | null;
+  project: unknown;
+}
+
+const previousHome = process.env.OPENCODEX_HOME;
+let home = "";
+let codexHome: IsolatedCodexHome | null = null;
+
+beforeEach(() => {
+  home = mkdtempSync(join(tmpdir(), "ocx-google-pool-request-"));
+  process.env.OPENCODEX_HOME = home;
+  codexHome = installIsolatedCodexHome("ocx-google-pool-request-codex-");
+  clearGoogleAntigravityAccountPoolState();
+  clearAccountQuotaCache("google-antigravity");
+  clearRequestLogsForTests();
+});
+
+afterEach(() => {
+  clearGoogleAntigravityAccountPoolState();
+  clearAccountQuotaCache("google-antigravity");
+  codexHome?.restore();
+  codexHome = null;
+  rmSync(home, { recursive: true, force: true });
+  if (previousHome === undefined) delete process.env.OPENCODEX_HOME;
+  else process.env.OPENCODEX_HOME = previousHome;
+});
+
+async function seedAccounts(eligibleCount = 2, configuredCount = eligibleCount) {
+  const expires = Date.now() + 3_600_000;
+  const suffixes = ["a", "b", "c", "d", "e"].slice(0, configuredCount);
+  for (const suffix of suffixes) {
+    await saveCredential("google-antigravity", {
+      access: `request-token-${suffix}`,
+      refresh: `request-refresh-${suffix}`,
+      expires,
+      accountId: `request-identity-${suffix}`,
+      email: `request-${suffix}@example.test`,
+      projectId: `request-project-${suffix}`,
+    });
+  }
+  const set = getAccountSet("google-antigravity")!;
+  const accountIds = suffixes.map(suffix =>
+    set.accounts.find(account => account.credential.accountId === `request-identity-${suffix}`)!.id
+  );
+  await setActiveAccount("google-antigravity", accountIds[0]!);
+  for (const accountId of accountIds.slice(eligibleCount)) {
+    await markAccountNeedsReauth("google-antigravity", accountId, true);
+  }
+  return {
+    aId: accountIds[0]!,
+    bId: accountIds[1]!,
+    accountIds: accountIds.slice(0, eligibleCount),
+  };
+}
+
+function config(baseUrl: string): OcxConfig {
+  return {
+    port: 0,
+    hostname: "127.0.0.1",
+    defaultProvider: "google-antigravity",
+    providers: {
+      "google-antigravity": {
+        adapter: "google",
+        authMode: "oauth",
+        googleMode: "cloud-code-assist",
+        baseUrl,
+        allowPrivateNetwork: true,
+        defaultModel: "gemini-3.7-flash",
+        models: ["gemini-3.7-flash"],
+      },
+    },
+    googleAntigravityAccountPool: { enabled: true, autoSwitchThreshold: 80 },
+  } as OcxConfig;
+}
+
+function completedCcaResponse(): Response {
+  return Response.json({
+    response: {
+      candidates: [{ content: { parts: [{ text: "pool success" }] }, finishReason: "STOP" }],
+      usageMetadata: { promptTokenCount: 2, candidatesTokenCount: 2 },
+    },
+  });
+}
+
+function ccaSse(...payloads: unknown[]): Response {
+  return new Response(
+    payloads.map(payload => `data: ${JSON.stringify(payload)}\n\n`).join(""),
+    { headers: { "content-type": "text/event-stream" } },
+  );
+}
+
+const quotaResponses = new Map([
+  ["Bearer request-token-a", { status: 429, retryAfter: "300" }],
+  ["Bearer request-token-b", { status: 429, retryAfter: "500" }],
+  ["Bearer request-token-c", { status: 429, retryAfter: "400" }],
+  ["Bearer request-token-d", { status: 402, retryAfter: "450" }],
+  ["Bearer request-token-e", { status: 429, retryAfter: "600" }],
+]);
+
+function exhaustedCcaResponse(authorization: string | null): Response {
+  const quota = authorization ? quotaResponses.get(authorization) : undefined;
+  if (!quota) throw new Error(`unexpected Antigravity authorization: ${authorization ?? "missing"}`);
+  const suffix = authorization.at(-1);
+  return Response.json(
+    {
+      error: {
+        status: "RESOURCE_EXHAUSTED",
+        message: `Quota exceeded for request-${suffix}@example.test, request-identity-${suffix}, request-project-${suffix}`,
+      },
+    },
+    { status: quota.status, headers: { "retry-after": quota.retryAfter } },
+  );
+}
+
+describe("google antigravity Responses account failover", () => {
+  test("snapshot failure does not commit initial session affinity or dispatch upstream", async () => {
+    await saveCredential("google-antigravity", {
+      access: "request-token-without-project",
+      refresh: "request-refresh-without-project",
+      expires: Date.now() + 3_600_000,
+      accountId: "request-identity-without-project",
+      email: "request-without-project@example.test",
+    });
+    let dispatches = 0;
+    const upstream = Bun.serve({
+      port: 0,
+      hostname: "127.0.0.1",
+      fetch() {
+        dispatches += 1;
+        return completedCcaResponse();
+      },
+    });
+    saveConfig(config(upstream.url.origin));
+    const proxy = startServer(0);
+    try {
+      expect(googleAntigravitySessionAffinitySizeForTests()).toBe(0);
+      const response = await fetch(new URL("/v1/responses", proxy.url), {
+        method: "POST",
+        headers: { "content-type": "application/json", "thread-id": "snapshot-failure" },
+        body: JSON.stringify({ model: "gemini-3.7-flash", input: "hello", stream: false }),
+      });
+      expect(response.status).toBe(401);
+      expect(dispatches).toBe(0);
+      expect(googleAntigravitySessionAffinitySizeForTests()).toBe(0);
+    } finally {
+      await proxy.stop(true);
+      upstream.stop(true);
+    }
+  });
+
+  test.each([1, 2, 3, 4])(
+    "initial exhaustion across %d eligible accounts returns all-cooled without an extra dispatch",
+    async eligibleCount => {
+    const { accountIds } = await seedAccounts(eligibleCount, 4);
+    const dispatches: Dispatch[] = [];
+    const upstream = Bun.serve({
+      port: 0,
+      hostname: "127.0.0.1",
+      async fetch(request) {
+        const body = await request.json() as { project?: unknown };
+        dispatches.push({
+          authorization: request.headers.get("authorization"),
+          project: body.project,
+        });
+        return exhaustedCcaResponse(request.headers.get("authorization"));
+      },
+    });
+    saveConfig(config(upstream.url.origin));
+    const proxy = startServer(0);
+    try {
+      const request = () => fetch(new URL("/v1/responses", proxy.url), {
+        method: "POST",
+        headers: { "content-type": "application/json", "thread-id": `initial-exhaustion-${eligibleCount}` },
+        body: JSON.stringify({ model: "gemini-3.7-flash", input: "hello", stream: false }),
+      });
+
+      const exhausted = await request();
+      const exhaustedBody = await exhausted.text();
+      expect([exhausted.status, exhausted.headers.get("retry-after")]).toEqual([429, "300"]);
+      expect(exhaustedBody).toContain("All Google Antigravity OAuth accounts are temporarily rate-limited");
+      expect(dispatches).toEqual(["a", "b", "c", "d"].slice(0, eligibleCount).map(suffix => ({
+        authorization: `Bearer request-token-${suffix}`,
+        project: `request-project-${suffix}`,
+      })));
+      expect(accountIds.every(id => getGoogleAntigravityAccountHealthSnapshot(id) !== null)).toBe(true);
+
+      const allCooled = await request();
+      const allCooledBody = await allCooled.text();
+      expect([allCooled.status, allCooled.headers.get("retry-after")]).toEqual([429, "300"]);
+      expect(allCooledBody).toBe(exhaustedBody);
+      expect(dispatches).toHaveLength(eligibleCount);
+    } finally {
+      await proxy.stop(true);
+      upstream.stop(true);
+    }
+    },
+  );
+
+  test("the failover cap cools the fourth account without dispatching an eligible fifth account", async () => {
+    const { accountIds } = await seedAccounts(5, 5);
+    const dispatches: Dispatch[] = [];
+    const upstream = Bun.serve({
+      port: 0,
+      hostname: "127.0.0.1",
+      async fetch(request) {
+        const body = await request.json() as { project?: unknown };
+        dispatches.push({
+          authorization: request.headers.get("authorization"),
+          project: body.project,
+        });
+        return exhaustedCcaResponse(request.headers.get("authorization"));
+      },
+    });
+    saveConfig(config(upstream.url.origin));
+    const proxy = startServer(0);
+    try {
+      const response = await fetch(new URL("/v1/responses", proxy.url), {
+        method: "POST",
+        headers: { "content-type": "application/json", "thread-id": "five-account-cap" },
+        body: JSON.stringify({ model: "gemini-3.7-flash", input: "hello", stream: false }),
+      });
+      const body = await response.text();
+
+      expect([response.status, response.headers.get("retry-after")]).toEqual([402, "450"]);
+      expect(body).toContain("Google Antigravity quota exhausted after bounded account failover");
+      for (const secret of [
+        "request-d@example.test",
+        "request-identity-d",
+        "request-project-d",
+      ]) expect(body).not.toContain(secret);
+      expect(dispatches).toEqual(["a", "b", "c", "d"].map(suffix => ({
+        authorization: `Bearer request-token-${suffix}`,
+        project: `request-project-${suffix}`,
+      })));
+      expect(accountIds.slice(0, 4).every(id => getGoogleAntigravityAccountHealthSnapshot(id) !== null)).toBe(true);
+      expect(getGoogleAntigravityAccountHealthSnapshot(accountIds[4]!)).toBeNull();
+    } finally {
+      await proxy.stop(true);
+      upstream.stop(true);
+    }
+  });
+
+  test.each([1, 2, 3, 4])(
+    "terminal continuation exhaustion across %d eligible accounts surfaces all-cooled",
+    async eligibleCount => {
+    const { accountIds } = await seedAccounts(eligibleCount, 4);
+    let initialSends = 0;
+    const quotaDispatches: Dispatch[] = [];
+    const upstream = Bun.serve({
+      port: 0,
+      hostname: "127.0.0.1",
+      async fetch(request) {
+        if (initialSends === 0) {
+          initialSends += 1;
+          return Response.json({
+            response: {
+              candidates: [{ content: { parts: [] }, finishReason: "STOP" }],
+              usageMetadata: { promptTokenCount: 1, candidatesTokenCount: 0 },
+            },
+          });
+        }
+        const body = await request.json() as { project?: unknown };
+        quotaDispatches.push({
+          authorization: request.headers.get("authorization"),
+          project: body.project,
+        });
+        return exhaustedCcaResponse(request.headers.get("authorization"));
+      },
+    });
+    saveConfig({ ...config(upstream.url.origin), emptyCompletionRetry: true });
+    const proxy = startServer(0);
+    try {
+      const first = await fetch(new URL("/v1/responses", proxy.url), {
+        method: "POST",
+        headers: { "content-type": "application/json", "thread-id": `continuation-exhaustion-${eligibleCount}` },
+        body: JSON.stringify({ model: "gemini-3.7-flash", input: "hello", stream: false }),
+      });
+      const firstBody = await first.text();
+      expect([first.status, first.headers.get("retry-after")]).toEqual([429, "300"]);
+      expect(firstBody).toContain("All Google Antigravity OAuth accounts are temporarily rate-limited");
+      expect(initialSends).toBe(1);
+      expect(quotaDispatches).toEqual(["a", "b", "c", "d"].slice(0, eligibleCount).map(suffix => ({
+        authorization: `Bearer request-token-${suffix}`,
+        project: `request-project-${suffix}`,
+      })));
+      expect(accountIds.every(id => getGoogleAntigravityAccountHealthSnapshot(id) !== null)).toBe(true);
+
+      const allCooled = await fetch(new URL("/v1/responses", proxy.url), {
+        method: "POST",
+        headers: { "content-type": "application/json", "thread-id": `continuation-exhaustion-${eligibleCount}` },
+        body: JSON.stringify({ model: "gemini-3.7-flash", input: "again", stream: false }),
+      });
+      const allCooledBody = await allCooled.text();
+      expect([allCooled.status, allCooled.headers.get("retry-after")]).toEqual([429, "300"]);
+      expect(allCooledBody).toBe(firstBody);
+      expect(initialSends).toBe(1);
+      expect(quotaDispatches).toHaveLength(eligibleCount);
+    } finally {
+      await proxy.stop(true);
+      upstream.stop(true);
+    }
+    },
+  );
+
+  test("terminal continuation cap preserves quota status without dispatching an eligible fifth account", async () => {
+    const { accountIds } = await seedAccounts(5, 5);
+    let initialSends = 0;
+    const quotaDispatches: Dispatch[] = [];
+    const upstream = Bun.serve({
+      port: 0,
+      hostname: "127.0.0.1",
+      async fetch(request) {
+        if (initialSends === 0) {
+          initialSends += 1;
+          return Response.json({
+            response: {
+              candidates: [{ content: { parts: [] }, finishReason: "STOP" }],
+              usageMetadata: { promptTokenCount: 1, candidatesTokenCount: 0 },
+            },
+          });
+        }
+        const body = await request.json() as { project?: unknown };
+        quotaDispatches.push({
+          authorization: request.headers.get("authorization"),
+          project: body.project,
+        });
+        return exhaustedCcaResponse(request.headers.get("authorization"));
+      },
+    });
+    saveConfig({ ...config(upstream.url.origin), emptyCompletionRetry: true });
+    const proxy = startServer(0);
+    try {
+      const response = await fetch(new URL("/v1/responses", proxy.url), {
+        method: "POST",
+        headers: { "content-type": "application/json", "thread-id": "five-account-continuation-cap" },
+        body: JSON.stringify({ model: "gemini-3.7-flash", input: "hello", stream: false }),
+      });
+      const body = await response.text();
+
+      expect([response.status, response.headers.get("retry-after")]).toEqual([402, "450"]);
+      expect(body).toContain("Google Antigravity quota exhausted after bounded account failover");
+      expect(initialSends).toBe(1);
+      expect(quotaDispatches).toEqual(["a", "b", "c", "d"].map(suffix => ({
+        authorization: `Bearer request-token-${suffix}`,
+        project: `request-project-${suffix}`,
+      })));
+      expect(accountIds.slice(0, 4).every(id => getGoogleAntigravityAccountHealthSnapshot(id) !== null)).toBe(true);
+      expect(getGoogleAntigravityAccountHealthSnapshot(accountIds[4]!)).toBeNull();
+    } finally {
+      await proxy.stop(true);
+      upstream.stop(true);
+    }
+  });
+
+  test("streaming quota failover completes before parsing the successful account stream", async () => {
+    await seedAccounts();
+    const dispatches: Dispatch[] = [];
+    const upstream = Bun.serve({
+      port: 0,
+      hostname: "127.0.0.1",
+      async fetch(request) {
+        const body = await request.json() as { project?: unknown };
+        const authorization = request.headers.get("authorization");
+        dispatches.push({ authorization, project: body.project });
+        if (authorization === "Bearer request-token-a") {
+          return Response.json({
+            error: { status: "RESOURCE_EXHAUSTED", message: "Quota exceeded for this account" },
+          }, { status: 429 });
+        }
+        return ccaSse({
+          response: {
+            candidates: [{ content: { parts: [{ text: "stream pool success" }] }, finishReason: "STOP" }],
+            usageMetadata: { promptTokenCount: 2, candidatesTokenCount: 3 },
+          },
+        });
+      },
+    });
+    saveConfig(config(upstream.url.origin));
+    const proxy = startServer(0);
+    try {
+      const response = await fetch(new URL("/v1/responses", proxy.url), {
+        method: "POST",
+        headers: { "content-type": "application/json", "thread-id": "stream-failover" },
+        body: JSON.stringify({ model: "gemini-3.7-flash", input: "hello", stream: true }),
+      });
+      expect(response.status).toBe(200);
+      expect(await response.text()).toContain("stream pool success");
+      expect(dispatches).toHaveLength(2);
+    } finally {
+      await proxy.stop(true);
+      upstream.stop(true);
+    }
+  });
+
+  test("a partial successful stream is never replayed onto another account", async () => {
+    await seedAccounts();
+    const dispatches: Dispatch[] = [];
+    const upstream = Bun.serve({
+      port: 0,
+      hostname: "127.0.0.1",
+      async fetch(request) {
+        const body = await request.json() as { project?: unknown };
+        dispatches.push({
+          authorization: request.headers.get("authorization"),
+          project: body.project,
+        });
+        return ccaSse(
+          { response: { candidates: [{ content: { parts: [{ text: "partial stream output" }] } }] } },
+          { error: { message: "stream failed after output" } },
+        );
+      },
+    });
+    saveConfig(config(upstream.url.origin));
+    const proxy = startServer(0);
+    try {
+      const response = await fetch(new URL("/v1/responses", proxy.url), {
+        method: "POST",
+        headers: { "content-type": "application/json", "thread-id": "partial-stream" },
+        body: JSON.stringify({ model: "gemini-3.7-flash", input: "hello", stream: true }),
+      });
+      const text = await response.text();
+      expect(text).toContain("partial stream output");
+      expect(text).toContain("stream failed after output");
+      expect(dispatches).toEqual([
+        { authorization: "Bearer request-token-a", project: "request-project-a" },
+      ]);
+    } finally {
+      await proxy.stop(true);
+      upstream.stop(true);
+    }
+  });
+
+  test.each([429, 402])(
+    "final %d cools the account and rebuilds token plus project for the next account",
+    async failureStatus => {
+    const { aId, bId } = await seedAccounts();
+    const dispatches: Dispatch[] = [];
+    const upstream = Bun.serve({
+      port: 0,
+      hostname: "127.0.0.1",
+      async fetch(request) {
+        const body = await request.json() as { project?: unknown };
+        const authorization = request.headers.get("authorization");
+        dispatches.push({ authorization, project: body.project });
+        if (authorization === "Bearer request-token-a") {
+          return Response.json({
+            error: { status: "RESOURCE_EXHAUSTED", message: "Quota exceeded for this account" },
+          }, { status: failureStatus, headers: { "retry-after": "45" } });
+        }
+        return completedCcaResponse();
+      },
+    });
+    saveConfig(config(upstream.url.origin));
+    const proxy = startServer(0);
+    try {
+      const response = await fetch(new URL("/v1/responses", proxy.url), {
+        method: "POST",
+        headers: { "content-type": "application/json", "thread-id": "google-pool-thread" },
+        body: JSON.stringify({ model: "gemini-3.7-flash", input: "hello", stream: false }),
+      });
+      expect(response.status).toBe(200);
+      expect(await response.text()).toContain("pool success");
+      expect(dispatches).toEqual([
+        { authorization: "Bearer request-token-a", project: "request-project-a" },
+        { authorization: "Bearer request-token-b", project: "request-project-b" },
+      ]);
+      expect(getAccountSet("google-antigravity")?.activeAccountId).toBe(bId);
+      expect(aId).not.toBe(bId);
+      expect(getGoogleAntigravityAccountHealthSnapshot(aId)).toMatchObject({
+        cooldownSource: "retry-after",
+        cooldownUntil: expect.any(Number),
+      });
+
+      // Even if an operator-visible active pointer changes between turns, the
+      // same thread stays on B until B itself becomes ineligible/cooled.
+      await setActiveAccount("google-antigravity", aId);
+      const sticky = await fetch(new URL("/v1/responses", proxy.url), {
+        method: "POST",
+        headers: { "content-type": "application/json", "thread-id": "google-pool-thread" },
+        body: JSON.stringify({ model: "gemini-3.7-flash", input: "again", stream: false }),
+      });
+      expect(sticky.status).toBe(200);
+      expect(dispatches.at(-1)).toEqual({
+        authorization: "Bearer request-token-b",
+        project: "request-project-b",
+      });
+      expect(dispatches).toHaveLength(3);
+
+      const logs = getRequestLogEntries();
+      const failoverLog = logs.find(entry => entry.attempts?.some(attempt =>
+        attempt.recoveryKinds.includes("google-antigravity-oauth-quota")
+      ));
+      expect(failoverLog).toMatchObject({
+        provider: formatGoogleAntigravityProviderForLog("google-antigravity", bId),
+        attempts: [{
+          provider: formatGoogleAntigravityProviderForLog("google-antigravity", bId),
+          sendCount: 2,
+          recoveryKinds: ["google-antigravity-oauth-quota"],
+        }],
+      });
+      const serializedLogs = JSON.stringify(logs);
+      for (const secret of [
+        aId,
+        bId,
+        "request-token-a",
+        "request-token-b",
+        "request-project-a",
+        "request-project-b",
+        "request-a@example.test",
+        "request-b@example.test",
+      ]) expect(serializedLogs).not.toContain(secret);
+    } finally {
+      await proxy.stop(true);
+      upstream.stop(true);
+    }
+    },
+  );
+});
diff --git a/tests/google-antigravity-account-pool.test.ts b/tests/google-antigravity-account-pool.test.ts
new file mode 100644
index 0000000000..e155a332cf
--- /dev/null
+++ b/tests/google-antigravity-account-pool.test.ts
@@ -0,0 +1,287 @@
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
+import { mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import {
+  bindGoogleAntigravitySessionAffinity,
+  clearGoogleAntigravityAccountPoolState,
+  formatGoogleAntigravityProviderForLog,
+  getEligibleGoogleAntigravityAccounts,
+  getGoogleAntigravityAccountHealthSnapshot,
+  getGoogleAntigravityPoolAccessSnapshot,
+  getGoogleAntigravityPoolRetryAfterSeconds,
+  googleAntigravityAutoSwitchThreshold,
+  googleAntigravitySessionAffinitySizeForTests,
+  googleAntigravityUsageScore,
+  isGoogleAntigravityAccountPoolEnabled,
+  resolveGoogleAntigravityAccountForSession,
+  rotateGoogleAntigravityAccountOnQuotaError,
+} from "../src/oauth/google-antigravity-routing";
+import {
+  clearAccountQuotaCache,
+  getCachedProviderAccountQuota,
+  setCachedProviderAccountQuotaForTests,
+} from "../src/providers/quota";
+import { getAccountSet, saveCredential, setActiveAccount } from "../src/oauth/store";
+import type { OcxConfig, OcxProviderConfig } from "../src/types";
+
+const previousHome = process.env.OPENCODEX_HOME;
+let home = "";
+
+function config(
+  enabled?: boolean,
+  threshold?: number,
+  strategy?: "quota" | "round-robin" | "fill-first",
+): OcxConfig {
+  const provider: OcxProviderConfig = {
+    adapter: "google",
+    authMode: "oauth",
+    baseUrl: "https://daily-cloudcode-pa.googleapis.com",
+    googleMode: "cloud-code-assist",
+  };
+  return {
+    port: 0,
+    defaultProvider: "google-antigravity",
+    providers: { "google-antigravity": provider },
+    ...(enabled === undefined && threshold === undefined && strategy === undefined
+      ? {}
+      : {
+          googleAntigravityAccountPool: {
+            ...(enabled === undefined ? {} : { enabled }),
+            ...(threshold === undefined ? {} : { autoSwitchThreshold: threshold }),
+            ...(strategy === undefined ? {} : { strategy }),
+          },
+        }),
+  } as OcxConfig;
+}
+
+async function seedTwoAccounts() {
+  const expires = Date.now() + 3_600_000;
+  await saveCredential("google-antigravity", {
+    access: "google-access-a",
+    refresh: "google-refresh-a",
+    expires,
+    accountId: "google-identity-a",
+    email: "a@example.test",
+    projectId: "project-a",
+  });
+  await saveCredential("google-antigravity", {
+    access: "google-access-b",
+    refresh: "google-refresh-b",
+    expires,
+    accountId: "google-identity-b",
+    email: "b@example.test",
+    projectId: "project-b",
+  });
+  const set = getAccountSet("google-antigravity")!;
+  const a = set.accounts.find(account => account.credential.accountId === "google-identity-a")!;
+  const b = set.accounts.find(account => account.credential.accountId === "google-identity-b")!;
+  await setActiveAccount("google-antigravity", a.id);
+  return { aId: a.id, bId: b.id };
+}
+
+beforeEach(() => {
+  home = mkdtempSync(join(tmpdir(), "ocx-google-antigravity-pool-"));
+  process.env.OPENCODEX_HOME = home;
+  clearGoogleAntigravityAccountPoolState();
+  clearAccountQuotaCache("google-antigravity");
+});
+
+afterEach(() => {
+  clearGoogleAntigravityAccountPoolState();
+  clearAccountQuotaCache("google-antigravity");
+  rmSync(home, { recursive: true, force: true });
+  if (previousHome === undefined) delete process.env.OPENCODEX_HOME;
+  else process.env.OPENCODEX_HOME = previousHome;
+});
+
+describe("google antigravity account pool", () => {
+  test("is backward-compatible and disabled by default", async () => {
+    const { aId } = await seedTwoAccounts();
+    expect(isGoogleAntigravityAccountPoolEnabled(config())).toBe(false);
+    expect(resolveGoogleAntigravityAccountForSession("thread-1", "gemini-3.7-flash", config())).toEqual({
+      accountId: aId,
+      reason: "pool-disabled",
+    });
+    expect(rotateGoogleAntigravityAccountOnQuotaError(
+      config(), aId, "30", "gemini-3.7-flash", "thread-1",
+    )).toEqual({ kind: "pool-disabled" });
+    expect(getGoogleAntigravityAccountHealthSnapshot(aId)).toBeNull();
+  });
+
+  test("invalid persisted pool values fail safely to disabled/default policy", async () => {
+    const { aId } = await seedTwoAccounts();
+    const invalid = config() as OcxConfig & { googleAntigravityAccountPool: Record<string, unknown> };
+    invalid.googleAntigravityAccountPool = {
+      enabled: "yes",
+      autoSwitchThreshold: 101,
+      strategy: "weighted",
+      stickyLimit: 0,
+    };
+    expect(isGoogleAntigravityAccountPoolEnabled(invalid)).toBe(false);
+    expect(googleAntigravityAutoSwitchThreshold(invalid)).toBe(80);
+    expect(resolveGoogleAntigravityAccountForSession("invalid", "gemini-3.7-flash", invalid)).toEqual({
+      accountId: aId,
+      reason: "pool-disabled",
+    });
+  });
+
+  test("switches above threshold using the maximum relevant custom window", async () => {
+    const { aId, bId } = await seedTwoAccounts();
+    setCachedProviderAccountQuotaForTests("google-antigravity", aId, {
+      customWindows: [
+        { label: "Gem", percent: 81 },
+        { label: "Gemini secondary", percent: 92 },
+        { label: "Cla", percent: 5 },
+      ],
+      updatedAt: Date.now(),
+    });
+    setCachedProviderAccountQuotaForTests("google-antigravity", bId, {
+      customWindows: [
+        { label: "Gem", percent: 20 },
+        { label: "Cla", percent: 97 },
+      ],
+      updatedAt: Date.now(),
+    });
+
+    expect(googleAntigravityUsageScore(
+      "gemini-3.7-flash",
+      getCachedProviderAccountQuota("google-antigravity", aId),
+    )).toBe(92);
+    expect(resolveGoogleAntigravityAccountForSession("thread-gemini", "gemini-3.7-flash", config(true, 80))).toEqual({
+      accountId: bId,
+      reason: "lowest-usage",
+    });
+    expect(resolveGoogleAntigravityAccountForSession("thread-claude", "claude-sonnet-4-6", config(true, 80))).toEqual({
+      accountId: aId,
+      reason: "active",
+    });
+  });
+
+  test("unknown usage holds the healthy active account", async () => {
+    const { aId, bId } = await seedTwoAccounts();
+    setCachedProviderAccountQuotaForTests("google-antigravity", bId, {
+      customWindows: [{ label: "Gem", percent: 4 }],
+      updatedAt: Date.now(),
+    });
+    expect(resolveGoogleAntigravityAccountForSession("unknown-active", "gemini-3.7-flash", config(true, 80))).toEqual({
+      accountId: aId,
+      reason: "active",
+    });
+  });
+
+  test("account resolution proposes without committing or refreshing session affinity", async () => {
+    const { aId } = await seedTwoAccounts();
+    const start = 1_800_000_000_000;
+
+    expect(resolveGoogleAntigravityAccountForSession(
+      "proposal-only", "gemini-3.7-flash", config(true, 80), start,
+    )).toMatchObject({ accountId: aId, reason: "active" });
+    expect(googleAntigravitySessionAffinitySizeForTests()).toBe(0);
+
+    bindGoogleAntigravitySessionAffinity("existing-affinity", aId, start);
+    expect(resolveGoogleAntigravityAccountForSession(
+      "existing-affinity", "gemini-3.7-flash", config(true, 80), start + 23 * 60 * 60_000,
+    )).toMatchObject({ accountId: aId, reason: "affinity" });
+    expect(googleAntigravitySessionAffinitySizeForTests()).toBe(1);
+
+    resolveGoogleAntigravityAccountForSession(
+      "expiry-probe", "gemini-3.7-flash", config(true, 80), start + 25 * 60 * 60_000,
+    );
+    expect(googleAntigravitySessionAffinitySizeForTests()).toBe(0);
+  });
+
+  test("round-robin honors stickyLimit and fill-first advances a drained active account", async () => {
+    const { aId, bId } = await seedTwoAccounts();
+    const roundRobin = config(true, 80, "round-robin");
+    roundRobin.googleAntigravityAccountPool!.stickyLimit = 2;
+    const picks = [
+      resolveGoogleAntigravityAccountForSession("rr-1", "gemini-3.7-flash", roundRobin).accountId,
+      resolveGoogleAntigravityAccountForSession("rr-2", "gemini-3.7-flash", roundRobin).accountId,
+      resolveGoogleAntigravityAccountForSession("rr-3", "gemini-3.7-flash", roundRobin).accountId,
+    ];
+    expect(picks[0]).toBe(picks[1]);
+    expect(picks[2]).not.toBe(picks[1]);
+
+    clearGoogleAntigravityAccountPoolState();
+    setCachedProviderAccountQuotaForTests("google-antigravity", aId, {
+      customWindows: [{ label: "Gem", percent: 90 }],
+      updatedAt: Date.now(),
+    });
+    setCachedProviderAccountQuotaForTests("google-antigravity", bId, {
+      customWindows: [{ label: "Gem", percent: 20 }],
+      updatedAt: Date.now(),
+    });
+    expect(resolveGoogleAntigravityAccountForSession(
+      "fill-1", "gemini-3.7-flash", config(true, 80, "fill-first"),
+    )).toMatchObject({ accountId: bId, reason: "fill-first" });
+  });
+
+  test("preserves thread affinity until the bound account is cooled", async () => {
+    const { aId, bId } = await seedTwoAccounts();
+    bindGoogleAntigravitySessionAffinity("sticky-thread", aId);
+    expect(resolveGoogleAntigravityAccountForSession("sticky-thread", "gemini-3.7-flash", config(true))).toMatchObject({
+      accountId: aId,
+      reason: "affinity",
+    });
+
+    expect(rotateGoogleAntigravityAccountOnQuotaError(
+      config(true), aId, "30", "gemini-3.7-flash", "sticky-thread",
+    )).toEqual({ kind: "next-account", accountId: bId });
+    expect(googleAntigravitySessionAffinitySizeForTests()).toBe(0);
+    bindGoogleAntigravitySessionAffinity("sticky-thread", bId);
+    expect(resolveGoogleAntigravityAccountForSession("sticky-thread", "gemini-3.7-flash", config(true))).toMatchObject({
+      accountId: bId,
+      reason: "affinity",
+    });
+  });
+
+  test("reports all-cooled with the earliest Retry-After", async () => {
+    const now = 1_800_000_000_000;
+    const { aId, bId } = await seedTwoAccounts();
+    expect(rotateGoogleAntigravityAccountOnQuotaError(
+      config(true), aId, "90", "gemini-3.7-flash", null, now,
+    )).toEqual({ kind: "next-account", accountId: bId });
+    expect(rotateGoogleAntigravityAccountOnQuotaError(
+      config(true), bId, "30", "gemini-3.7-flash", null, now,
+    )).toEqual({ kind: "all-cooled" });
+    expect(resolveGoogleAntigravityAccountForSession("all-cooled", "gemini-3.7-flash", config(true), now)).toEqual({
+      accountId: null,
+      reason: "all-cooled",
+    });
+    expect(getEligibleGoogleAntigravityAccounts(now)).toEqual([]);
+    expect(getGoogleAntigravityPoolRetryAfterSeconds(now)).toBe(30);
+  });
+
+  test("uses non-PII labels and bounds retained affinity state", async () => {
+    const { aId } = await seedTwoAccounts();
+    const label = formatGoogleAntigravityProviderForLog("google-antigravity", aId);
+    expect(label).toMatch(/^google-antigravity-p[a-f0-9]{6}$/);
+    expect(label).not.toContain(aId);
+
+    bindGoogleAntigravitySessionAffinity("x".repeat(513), aId);
+    expect(googleAntigravitySessionAffinitySizeForTests()).toBe(0);
+    for (let index = 0; index < 2_050; index++) {
+      bindGoogleAntigravitySessionAffinity(`thread-${index}`, aId, index);
+    }
+    expect(googleAntigravitySessionAffinitySizeForTests()).toBeLessThanOrEqual(2_000);
+  });
+
+  test("resolves access token and projectId from the same selected account snapshot", async () => {
+    const { aId, bId } = await seedTwoAccounts();
+    expect(getAccountSet("google-antigravity")?.activeAccountId).toBe(aId);
+
+    const snapshot = await getGoogleAntigravityPoolAccessSnapshot(bId);
+
+    expect(snapshot).toMatchObject({
+      provider: "google-antigravity",
+      accountId: bId,
+      accessToken: "google-access-b",
+      projectId: "project-b",
+      generation: expect.any(String),
+    });
+    expect(snapshot.accessToken).not.toBe("google-access-a");
+    expect(snapshot.projectId).not.toBe("project-a");
+    expect(getAccountSet("google-antigravity")?.activeAccountId).toBe(aId);
+  });
+});
diff --git a/tests/provider-account-quota.test.ts b/tests/provider-account-quota.test.ts
index 634c663ca2..7dba97cc46 100644
--- a/tests/provider-account-quota.test.ts
+++ b/tests/provider-account-quota.test.ts
@@ -29,6 +29,24 @@ async function seedTwoAccounts(): Promise<void> {
   await saveCredential("anthropic", { access: "token-second", refresh: "refresh-second", expires, ...SECOND });
 }
 
+async function seedTwoGoogleAccounts(): Promise<void> {
+  const expires = Date.now() + 60 * 60_000;
+  await saveCredential("google-antigravity", {
+    access: "google-token-first",
+    refresh: "google-refresh-first",
+    expires,
+    accountId: "google-account-first",
+    projectId: "google-project-first",
+  });
+  await saveCredential("google-antigravity", {
+    access: "google-token-second",
+    refresh: "google-refresh-second",
+    expires,
+    accountId: "google-account-second",
+    projectId: "google-project-second",
+  });
+}
+
 function usageBody(fiveHour: number, sevenDay: number): string {
   return JSON.stringify({
     five_hour: { utilization: fiveHour, resets_at: "2026-07-05T12:00:00Z" },
@@ -54,6 +72,49 @@ afterEach(() => {
 });
 
 describe("fetchProviderAccountQuotas", () => {
+  test("Google Antigravity probes each account with its own token and project snapshot", async () => {
+    await seedTwoGoogleAccounts();
+    const seen: Array<{ authorization: string | null; project: unknown }> = [];
+    globalThis.fetch = (async (input: RequestInfo | URL, init?: RequestInit) => {
+      expect(String(input)).toBe("https://daily-cloudcode-pa.googleapis.com/v1internal:fetchAvailableModels");
+      const authorization = new Headers(init?.headers).get("authorization");
+      const body = JSON.parse(String(init?.body)) as { project?: unknown };
+      seen.push({ authorization, project: body.project });
+      const gemRemaining = authorization === "Bearer google-token-first" ? 0.1 : 0.8;
+      return Response.json({
+        models: {
+          "gemini-3.7-flash": {
+            displayName: "Gemini 3.7 Flash",
+            quotaInfo: { remainingFraction: gemRemaining, resetTime: "2026-09-01T00:00:00Z" },
+          },
+          "claude-sonnet-4-6": {
+            displayName: "Claude Sonnet 4.6",
+            quotaInfo: { remainingFraction: 0.5, resetTime: "2026-09-01T01:00:00Z" },
+          },
+        },
+      });
+    }) as typeof fetch;
+
+    expect(supportsPerAccountQuota("google-antigravity")).toBe(true);
+    const rows = await fetchProviderAccountQuotas("google-antigravity");
+    expect(rows.map(row => row.quota?.customWindows).sort((a, b) =>
+      (a?.[0]?.percent ?? 0) - (b?.[0]?.percent ?? 0)
+    )).toEqual([
+      [
+        { label: "Gem", percent: 20, resetAt: Date.parse("2026-09-01T00:00:00Z") },
+        { label: "Cla", percent: 50, resetAt: Date.parse("2026-09-01T01:00:00Z") },
+      ],
+      [
+        { label: "Gem", percent: 90, resetAt: Date.parse("2026-09-01T00:00:00Z") },
+        { label: "Cla", percent: 50, resetAt: Date.parse("2026-09-01T01:00:00Z") },
+      ],
+    ]);
+    expect(seen).toEqual(expect.arrayContaining([
+      { authorization: "Bearer google-token-first", project: "google-project-first" },
+      { authorization: "Bearer google-token-second", project: "google-project-second" },
+    ]));
+  });
+
   test("reports each account's own rate limits, keyed by the account's bearer token", async () => {
     await seedTwoAccounts();
     const seenTokens: string[] = [];

From 1315ef0ca8fccb2bd73d3cdda3483ce2229f25b5 Mon Sep 17 00:00:00 2001
From: roy6732856 <roy6732856@gmail.com>
Date: Wed, 26 Aug 2026 12:58:38 +0800
Subject: [PATCH 2/5] fix(google): reconcile OAuth failover ownership

---
 .../fr/reference/cli/providers-accounts.md    |   5 +
 .../fr/reference/configuration/providers.md   |   9 ++
 .../docs/fr/reference/management-api.md       |  13 +-
 .../ja/reference/cli/providers-accounts.md    |   4 +
 .../ja/reference/configuration/providers.md   |  65 ++++++++++
 .../docs/ja/reference/management-api.md       |  11 +-
 .../ko/reference/cli/providers-accounts.md    |   4 +
 .../ko/reference/configuration/providers.md   |   9 ++
 .../docs/ko/reference/management-api.md       |  13 +-
 .../docs/reference/cli/providers-accounts.md  |   5 +
 .../docs/reference/configuration/providers.md |  12 +-
 .../content/docs/reference/management-api.md  |  11 +-
 .../ru/reference/cli/providers-accounts.md    |   5 +
 .../ru/reference/configuration/providers.md   |  10 ++
 .../docs/ru/reference/management-api.md       |  11 +-
 .../tr/reference/cli/providers-accounts.md    |   5 +
 .../tr/reference/configuration/providers.md   |  66 +++++++++++
 .../docs/tr/reference/management-api.md       |  11 +-
 .../zh-cn/reference/cli/providers-accounts.md |   4 +
 .../reference/configuration/providers.md      |  57 +++++++++
 .../docs/zh-cn/reference/management-api.md    |   8 +-
 .../zh-tw/reference/cli/providers-accounts.md |   7 +-
 .../reference/configuration/providers.md      |  12 ++
 .../docs/zh-tw/reference/management-api.md    |   8 +-
 src/oauth/generic-account-failover.ts         |   6 +
 src/server/management/oauth-account-routes.ts |  24 +++-
 src/server/responses/core.ts                  |  24 ++--
 src/types/config.ts                           |   4 +-
 tests/account-pool-management-api.test.ts     | 111 ++++++++++++++++--
 tests/config.test.ts                          |  34 ++++++
 tests/generic-oauth-failover.test.ts          |  13 +-
 tests/state-store-sweeper.test.ts             |   1 +
 32 files changed, 529 insertions(+), 53 deletions(-)

diff --git a/docs-site/src/content/docs/fr/reference/cli/providers-accounts.md b/docs-site/src/content/docs/fr/reference/cli/providers-accounts.md
index f993c59074..fa7ee3fec1 100644
--- a/docs-site/src/content/docs/fr/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/fr/reference/cli/providers-accounts.md
@@ -216,6 +216,11 @@ activent ou désactivent le pool tout en conservant son seuil. `threshold <n>` e
 Les fournisseurs non pris en charge et les valeurs invalides entraînent le code de sortie 1.
 `--json` renvoie :
 
+Pour `google-antigravity`, `off` désactive le pool spécialisé tenant compte des quotas et enregistre
+également la désactivation du basculement OAuth générique sur 429 pour ce fournisseur ; la commande
+passe donc réellement en mode strict à compte unique. Définir directement uniquement
+`googleAntigravityAccountPool.enabled: false` ne modifie pas cette stratégie générique distincte.
+
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
diff --git a/docs-site/src/content/docs/fr/reference/configuration/providers.md b/docs-site/src/content/docs/fr/reference/configuration/providers.md
index 9bb593b002..e55b7c3195 100644
--- a/docs-site/src/content/docs/fr/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/fr/reference/configuration/providers.md
@@ -84,6 +84,7 @@ sauvegarde dont le contenu diffère, puis réécrit en identifiants sans préfix
 | `modelContextWindows?` | `Record<string, number>` | Valeurs de repli ou plafonds de contexte par modèle. Ils remplacent `contextWindow` : une fenêtre inconnue utilise la valeur configurée, tandis que des métadonnées actives plus faibles restent déterminantes. |
 | `modelInputModalities?` | `Record<string, string[]>` | Conseils de saisie par modèle tels que `["text"]` ou `["text", "image"]`. |
 | `modelMaxInputTokens?` | `Record<string, number>` | Limites d'entrée maximales positives par modèle utilisées pour les conseils de compactage automatique du catalogue. |
+| `modelAutoCompactTokenLimits?` | `Record<string, number>` | Budgets souples de compactage automatique par modèle, sous forme d'entiers sûrs positifs. Ils peuvent uniquement abaisser l'enveloppe effective de 90 % du contexte ou de l'entrée maximale et sont omis lorsqu'aucune fenêtre de contexte faisant autorité n'est connue. Pour le fournisseur canonique `openai`, les clés doivent être les identifiants exacts de modèles natifs pris en charge, sans préfixe de fournisseur ni de sélecteur de compte. PATCH fusionne les entrées ; `null` supprime une clé, tandis que `null` pour le champ entier efface la table. Ces marqueurs `null` sont réservés à PATCH. |
 | `defaultMaxOutputTokens?` | `number` | Solution de secours `openai-chat` à l’échelle du fournisseur lorsque le client omet `max_output_tokens`. |
 | `modelMaxOutputTokens?` | `Record<string, number>` | Budgets de repli `openai-chat` positifs par modèle ; les correspondances exactes ou par motif priment sur la valeur par défaut du fournisseur. |
 | `modelCosts?` | `Record<string, Cost4>` | Prix affichés par modèle (USD par 1M de jetons), indexés par l'identifiant exact du modèle en amont de ce fournisseur — et non par un identifiant de fournisseur ni par une étiquette routée `provider/model`, par exemple `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`. Tout identifiant de modèle constitue une clé valide : les fournisseurs personnalisés peuvent cibler n'importe quel point de terminaison compatible avec OpenAI au moyen de l'adaptateur `openai-chat`, et les identifiants de fournisseur locaux ou internes fonctionnent même s'ils sont absents des catalogues intégrés. Les prix configurés par l'utilisateur priment sur les catalogues intégrés dans les estimations des pages Journaux (`~$`) et Utilisation. Les entrées historiques sont recalculées à partir de la surcharge actuelle ; modifier un prix peut donc changer les totaux antérieurs. L'ordre de repli est le suivant : `modelCosts` défini par l'utilisateur → catalogue jawcode → surcharge des prix attendus → repli propre au fournisseur au niveau du modèle. Une entrée entièrement nulle passe à la source suivante. Chaque tarif doit être un nombre fini positif ou nul, inférieur ou égal à 1 000 000 (USD par 1M de jetons) ; les lignes hors plage sont rejetées par l'interface de gestion et ignorées au chargement. Ces valeurs servent uniquement à l'estimation lors de l'affichage : les surcharges n'affectent jamais le routage, la sélection des comptes, les quotas ni la facturation. |
@@ -212,6 +213,14 @@ Cette option regroupe des comptes OAuth Google Antigravity uniquement pour le fo
 Assist `google-antigravity`. Elle est désactivée par défaut. Le pool ne fournit jamais d'identifiants à
 Google AI Studio, Vertex AI, une autre entrée de fournisseur ou une route utilisant une clé API.
 
+Désactiver ce pool spécialisé désactive la sélection selon les quotas, l'affinité de session et son
+budget de nouvelles tentatives après un 402 ou un 429. Cela ne désactive pas le mécanisme distinct de
+basculement OAuth générique sur 429, activé en fonction de la présence de comptes. Lorsque
+`googleAntigravityAccountPool.enabled` vaut `true`, le pool spécialisé prend en charge le routage
+Google ; lorsqu'il est désactivé, la stratégie générique peut encore assurer un basculement d'urgence
+après un 429. Pour limiter strictement Google à un seul compte, définissez également
+`providers.google-antigravity.oauthAccountFailover.enabled` sur `false`.
+
 | Clé | Type | Par défaut | Description |
 | --- | --- | --- | --- |
 | `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | Active l'affinité de session locale au processus et un basculement borné après une réponse finale 429 ou une réponse 402 remontée. |
diff --git a/docs-site/src/content/docs/fr/reference/management-api.md b/docs-site/src/content/docs/fr/reference/management-api.md
index 61be7f9c81..f1f851fa7e 100644
--- a/docs-site/src/content/docs/fr/reference/management-api.md
+++ b/docs-site/src/content/docs/fr/reference/management-api.md
@@ -197,11 +197,20 @@ conservent leur valeur actuelle ou par défaut. En cas de succès, la réponse c
 sans enveloppe `config`. Un corps mal formé ou qui n'est pas un objet, un fournisseur non pris en
 charge ou un champ invalide renvoie une réponse d'erreur HTTP 400 générique.
 
+Pour `google-antigravity`, écrire `enabled: false` enregistre également
+`providers.google-antigravity.oauthAccountFailover.enabled: false`. Cela garantit que le sens de
+« off » pour l'API de gestion et la CLI reste fidèle, en désactivant à la fois le pool spécialisé
+tenant compte des quotas et le mécanisme de basculement générique sur 429, autrement activé selon la
+présence de comptes. Modifier directement uniquement `googleAntigravityAccountPool.enabled` ne
+change pas cette stratégie générique.
+
 `POST /api/oauth/accounts/clear-cooldown` accepte `{ provider, accountId }`. Le fournisseur doit être
 `anthropic` ou `google-antigravity`, et le compte doit exister pour ce fournisseur. Un compte inconnu
 renvoie l'erreur HTTP 400 générique `account not found`. Un compte connu qui n'est pas en
-temporisation produit un succès idempotent : HTTP 200 avec `{ ok: true, cleared: false }`. Ces
-réponses ne contiennent jamais de jetons, d'adresses e-mail ni de clés complètes.
+temporisation produit un succès idempotent : HTTP 200 avec `{ ok: true, cleared: false }`. Pour
+Google, l'opération efface à la fois l'état du pool spécialisé et celui du mécanisme de basculement
+générique ; elle reste donc valable après un changement de mode du pool. Ces réponses ne contiennent
+jamais de jetons, d'adresses e-mail ni de clés complètes.
 
 Les réponses qui répertorient les identifiants sont délibérément masquées. Les jetons d'accès OAuth et les clés API complètes des
 fournisseurs ne sont pas renvoyés aux clients du tableau de bord.
diff --git a/docs-site/src/content/docs/ja/reference/cli/providers-accounts.md b/docs-site/src/content/docs/ja/reference/cli/providers-accounts.md
index f50cafbd70..ab5f3d0bb4 100644
--- a/docs-site/src/content/docs/ja/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/ja/reference/cli/providers-accounts.md
@@ -162,6 +162,10 @@ OAuth プロバイダーと API キー プロバイダーの場合、これに
 整数を保存し、0 以外ならプールを有効にします。`status` は現在のポリシーを読み取ります。未対応の
 プロバイダーと無効な値は終了コード 1 になります。`--json` は次を返します。
 
+`google-antigravity` では、`off` は専用のクォータ認識プールを無効にするだけでなく、プロバイダー単位の
+汎用 OAuth 429 オプトアウトも保存するため、コマンドは実際に厳格な単一アカウントモードへ移行します。
+`googleAntigravityAccountPool.enabled: false` だけを直接設定しても、この別個の汎用ポリシーは変更されません。
+
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
diff --git a/docs-site/src/content/docs/ja/reference/configuration/providers.md b/docs-site/src/content/docs/ja/reference/configuration/providers.md
index e7122dd2f2..3da7f62157 100644
--- a/docs-site/src/content/docs/ja/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ja/reference/configuration/providers.md
@@ -72,6 +72,7 @@ account を削除しても mapping は保持され、同じ id を再追加す
 | `modelContextWindows?` | `Record<string, number>` | モデルごとのコンテキスト値および上限。`contextWindow` より優先され、ウィンドウが不明なら設定値を使い、より小さいライブメタデータがあればそちらが優先されます。 |
 | `modelInputModalities?` | `Record<string, string[]>` | `["text"]` や `["text", "image"]` などのモデルごとの入力ヒント。 |
 | `modelMaxInputTokens?` | `Record<string, number>` |カタログの自動圧縮ヒントに使用されるモデルごとの正の最大入力制限。 |
+| `modelAutoCompactTokenLimits?` | `Record<string, number>` | モデルごとの正の安全な整数によるソフト自動圧縮予算。実効値であるコンテキストまたは最大入力の 90% の上限を下げることだけができ、信頼できるコンテキストウィンドウが不明な場合は出力されません。canonical `openai` では、キーは provider や account-selector の接頭辞を含まない、サポート対象の正確なネイティブモデル ID でなければなりません。provider PATCH はエントリをマージし、キーを `null` にするとそのキーを削除し、フィールド全体を `null` にするとマップを消去します。これらの `null` tombstone は PATCH 専用です。 |
 | `defaultMaxOutputTokens?` | `number` |クライアントが `max_output_tokens` を省略した場合の、プロバイダー全体の `openai-chat` フォールバック。 |
 | `modelMaxOutputTokens?` | `Record<string, number>` |モデルごとの `openai-chat` フォールバック バジェットがプラスになります。正確な/パターン一致はプロバイダーのデフォルトを上回ります。 |
 | `modelCosts?` | `Record<string, Cost4>` | モデルごとの表示価格（100万トークンあたりの米ドル）。そのプロバイダーの正確なアップストリーム モデル ID をキーにします（プロバイダー識別子やルーティングされた `provider/model` ラベルではありません）。値は `input`, `output`, `cacheRead`, `cacheWrite` の 4 フィールドです（例: `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`）。組み込みカタログにないモデル ID も、任意の OpenAI 互換エンドポイントを対象とするカスタムプロバイダーや、ローカル・内部プロバイダーで有効です。ユーザー設定の価格は Logs の `~$` と Usage の見積もりで組み込みカタログより優先されます。過去のエントリも現在のオーバーレイで再計算されるため、価格を編集すると過去の合計が変わることがあります（フォールバック順: ユーザー設定 → jawcode カタログ → expected-price オーバーレイ → モデル別ベンダー価格）。全ゼロのエントリは次のソースにフォールバックします。各レートは 0 以上の有限数で、最大 1,000,000（100万トークンあたりの米ドル）です。範囲外の行は管理境界で拒否され、読み込み時に破棄されます。表示専用の見積もりであり、ルーティング・アカウント選択・クォータ・請求には影響しません。 |
@@ -174,6 +175,11 @@ Anthropic アカウント ポリシーのリスクを理解していない限り
 Antigravity OAuth アカウントをプールします。デフォルトでは無効です。このプールの認証情報が
 Google AI Studio、Vertex AI、別のプロバイダー項目、API キールートに使われることはありません。
 
+この専用プールを無効にすると、クォータを考慮した選択、セッション親和性、および専用の 402/429
+再試行予算が無効になります。ただし、アカウントの存在によって有効になる別個の汎用 OAuth 429
+フォールバックは無効になりません。Google を厳密に単一アカウントで運用するには、
+`providers.google-antigravity.oauthAccountFailover.enabled` も `false` に設定してください。
+
 | キー | 型 | デフォルト | 説明 |
 | --- | --- | --- | --- |
 | `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | プロセスローカルなセッション親和性と、最終的な 429 または表面化した 402 に対する回数制限付きフェイルオーバーを有効にします。 |
@@ -201,6 +207,65 @@ Google AI Studio、Vertex AI、別のプロバイダー項目、API キールー
 受け入れない場合は無効のままにしてください。
 :::
 
+### `oauthAccountFailover`
+
+プロバイダー固有のプールが有効でない間、OAuth プロバイダーのアカウントがレート制限を受けると、
+同じプロバイダーの別のログイン済みアカウントへローテーションします。対象は xAI、Cursor、Kimi、
+GitHub Copilot、Google Antigravity、Nous です。`googleAntigravityAccountPool.enabled` が `true` の
+場合は、その専用プールが代わりに Google のルーティングを専有します。無効の場合でも、この汎用
+ポリシーによって緊急時の 429 フェイルオーバーが行われることがあります。
+
+**2 つ目のアカウントへのログインが、この機能を有効にする契機です。** 設定がない場合、対象
+プロバイダーに再認証が必要とされていないアカウントが 2 つ以上あるとローテーションが有効になります。
+これは `apiKeyPool` が 2 つ以上のキーを持つプールにすでに適用しているものと同じルールです。保存済み
+アカウントが 1 つのプロバイダーは、従来とまったく同じように動作します。
+
+| キー | 型 | デフォルト | 説明 |
+| --- | --- | --- | --- |
+| `oauthAccountFailover.enabled?` | `boolean` | アカウントの存在に応じる | グローバルなオーバーライドです。`false` はすべてのプロバイダーを単一アカウント動作に固定し、`true` はローテーションを強制的に有効にします。 |
+| `providers.<name>.oauthAccountFailover.enabled?` | `boolean` | 継承 | プロバイダー単位のオーバーライドです。グローバル設定とアカウントの存在より優先されます。 |
+
+利用規約上のリスクを避けたい特定のプロバイダーを厳密な単一アカウント動作に保つには、次のようにします。
+
+```json
+{
+  "providers": {
+    "cursor": {
+      "oauthAccountFailover": { "enabled": false }
+    }
+  }
+}
+```
+
+この設定は、ログイン、アカウントの追加、再認証を行っても保持されます。
+
+この機能は意図的に `anthropicAccountPool` より限定されています。セッション親和性、クォータ順位による
+選択、プローブリースはありません。答えるのは 1 つの問いだけです。つまり、429 を返したばかりの
+アカウントがクールダウン中のとき、利用できる別のアカウントがあるかどうかです。
+
+Codex プールと Anthropic プールは対象外であり、それぞれ独自のローテーションを維持します。この機能を
+有効にしても、どちらにも変更はありません。保存済みアカウントが 1 つだけのプロバイダーでは何も行われず、
+そのアカウントにクールダウンは記録されません。
+
+429 が発生すると、失敗したアカウントは `Retry-After` がある場合はその値（上限 15 分）を使い、ない場合は
+デフォルトのバックオフでクールダウンされます。その後、次の適格なアカウントでリクエストが再実行され、
+1 リクエストにつき最大 3 回ローテーションします。再認証が必要とマークされたアカウントは選択されません。
+クールダウンはプロセスローカルであるため、再起動すると失われます。
+
+ローテーションでは、代替アカウントのベアラーだけでなく、認証情報の**完全な**スナップショットを引き継ぎます。
+そのため、トークンとルーティングメタデータを対応付けるプロバイダー（たとえば Antigravity の Cloud Code Assist
+プロジェクト ID）で、あるアカウントのトークンを別のアカウントのメタデータと組み合わせて送信することはありません。
+
+現在の対象範囲は通常の Responses リクエストパスです。Cursor はレート制限を HTTP ステータスではなく
+アダプターイベントとして報告し、スタンドアロンの Antigravity 画像エンドポイントには独自のリクエストパスが
+あります。現時点では、どちらもローテーションしません。
+
+:::caution[実験的]
+サブスクリプションアカウント間のローテーションは 2 つ目のアカウントのクォータを消費し、一部のプロバイダーの
+利用規約に違反する可能性があります。このトレードオフを望まない場合は、グローバルまたは該当プロバイダーに対して
+`enabled: false` を設定してください。
+:::
+
 ### 管理されたレコードの形状
 
 `apiKeys[]` エントリには、`id`、`name`、生成された `key`、および ISO `createdAt` 文字列が含まれます。 `codexAccounts[]` エントリには `id`、`email`、および `isMain` が必要で、オプションの `plan`、`chatgptAccountId`、およびプライバシー セーフな `logLabel` が必要です。これらのレコードは通常、ダッシュボードで管理されます。
diff --git a/docs-site/src/content/docs/ja/reference/management-api.md b/docs-site/src/content/docs/ja/reference/management-api.md
index 8a46882aa6..e607b0a748 100644
--- a/docs-site/src/content/docs/ja/reference/management-api.md
+++ b/docs-site/src/content/docs/ja/reference/management-api.md
@@ -172,11 +172,18 @@ Authorization: Bearer <admin-token>
 stickyLimit, experimental: true }` を返し、`config` ラッパーはありません。不正形式またはオブジェクト
 以外の body、未対応の provider、不正なフィールドは、汎用の HTTP 400 エラーレスポンスになります。
 
+`google-antigravity` では、`enabled: false` を書き込むと
+`providers.google-antigravity.oauthAccountFailover.enabled: false` も保存されます。これにより、管理 API / CLI
+における「off」は、専用のクォータ認識プールと、アカウントの存在によって有効になる汎用 429
+フォールバックの両方を無効にするという実際の意味を保ちます。`googleAntigravityAccountPool.enabled` だけを
+直接編集しても、汎用ポリシーは変更されません。
+
 `POST /api/oauth/accounts/clear-cooldown` の body は `{ provider, accountId }` です。provider は
 `anthropic` または `google-antigravity` で、その account は同じ provider 内に存在する必要があります。
 不明な account は汎用の HTTP 400 `account not found` エラーになります。既知の account が cooldown
-中でなくても、HTTP 200 と `{ ok: true, cleared: false }` を返す冪等な成功です。これらのレスポンスには
-token、メールアドレス、完全な key は含まれません。
+中でなくても、HTTP 200 と `{ ok: true, cleared: false }` を返す冪等な成功です。Google では、この操作に
+よって専用プールの状態と汎用フォールバックの状態の両方がクリアされるため、プールモードの変更後も
+有効です。これらのレスポンスには token、メールアドレス、完全な key は含まれません。
 
 資格情報リストの応答は意図的にマスクされます。 OAuth アクセス トークンと完全なプロバイダー API キーはダッシュボード クライアントに返されません。
 
diff --git a/docs-site/src/content/docs/ko/reference/cli/providers-accounts.md b/docs-site/src/content/docs/ko/reference/cli/providers-accounts.md
index c2b8f5e7e5..24ee5d51e3 100644
--- a/docs-site/src/content/docs/ko/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/ko/reference/cli/providers-accounts.md
@@ -162,6 +162,10 @@ OAuth 및 API 키 제공자에는 제공자의 할당량 보고 엔드포인트
 정수를 저장하며, 0이 아니면 풀을 활성화합니다. `status`는 현재 정책을 읽습니다. 지원되지 않는
 제공자와 잘못된 값은 종료 코드 1로 끝납니다. `--json`은 다음을 반환합니다:
 
+`google-antigravity`에서 `off`는 특수 할당량 인식 풀을 비활성화하고 공급자별 일반 OAuth 429
+옵트아웃도 저장하므로, 이 명령은 실제로 엄격한 단일 계정 모드로 전환합니다.
+`googleAntigravityAccountPool.enabled: false`만 직접 설정하면 별도의 일반 정책은 변경되지 않습니다.
+
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
diff --git a/docs-site/src/content/docs/ko/reference/configuration/providers.md b/docs-site/src/content/docs/ko/reference/configuration/providers.md
index 2edd267ea4..61eaf5dbae 100644
--- a/docs-site/src/content/docs/ko/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ko/reference/configuration/providers.md
@@ -72,6 +72,7 @@ managed map을 활성화하면 privacy-safe selector를 만들고, 이후 계정
 | `modelContextWindows?` | `Record<string, number>` | 모델별 컨텍스트 값이자 상한입니다. `contextWindow`보다 우선하며, 창 크기를 알 수 없으면 설정값을 쓰고 더 작은 라이브 메타데이터가 있으면 그쪽을 따릅니다. |
 | `modelInputModalities?` | `Record<string, string[]>` | `["text"]` 또는 `["text", "image"]` 같은 모델별 입력 힌트입니다. |
 | `modelMaxInputTokens?` | `Record<string, number>` | 카탈로그 자동 압축 힌트에 쓰는 양수 모델별 최대 입력 한도입니다. |
+| `modelAutoCompactTokenLimits?` | `Record<string, number>` | 모델별 양의 안전 정수형 소프트 자동 압축 예산입니다. 유효한 컨텍스트 또는 최대 입력의 90% 한도를 낮출 수만 있으며, 신뢰할 수 있는 컨텍스트 창을 알 수 없으면 내보내지 않습니다. canonical `openai`에서는 키가 공급자나 계정 선택자 접두사가 없는 정확한 지원 네이티브 모델 ID여야 합니다. 공급자 PATCH는 항목을 병합하며, 키를 `null`로 지정하면 해당 키를 삭제하고 필드 전체를 `null`로 지정하면 맵을 지웁니다. 이 `null` tombstone은 PATCH에서만 사용할 수 있습니다. |
 | `defaultMaxOutputTokens?` | `number` | 클라이언트가 `max_output_tokens`를 생략했을 때 쓰는 공급자 전반의 `openai-chat` 폴백입니다. |
 | `modelMaxOutputTokens?` | `Record<string, number>` | 양수 모델별 `openai-chat` 폴백 예산입니다. 정확한 일치와 패턴 일치가 공급자 기본값보다 우선합니다. |
 | `modelCosts?` | `Record<string, Cost4>` | 모델별 표시 가격(100만 토큰당 USD). 해당 공급자의 정확한 업스트림 모델 ID를 키로 사용하며(공급자 식별자나 라우팅된 `provider/model` 레이블이 아님) 값은 `input`, `output`, `cacheRead`, `cacheWrite` 네 필드입니다(예: `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`). 커스텀 공급자는 `openai-chat` 어댑터로 임의의 OpenAI 호환 엔드포인트를 대상으로 할 수 있으며, 내장 카탈로그에 없는 로컬·내부 공급자 ID도 유효합니다. 사용자 구성 가격은 Logs `~$` 및 Usage 추정에서 내장 카탈로그보다 우선합니다. 기존 항목도 현재 오버레이로 다시 계산되므로 가격을 편집하면 과거 합계가 바뀔 수 있습니다(폴백 순서: 사용자 설정 → jawcode 카탈로그 → expected-price 오버레이 → 모델별 벤더 가격). 전부 0인 항목은 다음 소스로 폴백합니다. 각 요율은 0 이상의 유한한 숫자이며 최대 1,000,000(100만 토큰당 USD)입니다. 범위를 벗어난 행은 관리 경계에서 거부되고 로드 시 삭제됩니다. 표시 전용 추정이며 라우팅·계정 선택·할당량·청구에는 영향을 주지 않습니다. |
@@ -178,6 +179,14 @@ Anthropic 계정 정책 위험을 이해하지 못한다면 이 기능은 꺼두
 계정을 풀로 묶습니다. 기본값은 비활성화입니다. 이 풀의 자격 증명은 Google AI Studio,
 Vertex AI, 다른 공급자 항목 또는 API 키 경로에 사용되지 않습니다.
 
+이 특수 풀을 비활성화하면 할당량 인식 선택, 세션 affinity, 해당 풀의 402/429 재시도
+예산이 꺼집니다. 그러나 별도로 계정 존재 여부에 따라 활성화되는 일반 OAuth 429 fallback은
+비활성화되지 않습니다. Google을 엄격하게 단일 계정으로 유지하려면
+`providers.google-antigravity.oauthAccountFailover.enabled`도 `false`로 설정하십시오.
+
+`googleAntigravityAccountPool.enabled`가 `true`이면 이 특수 풀이 Google 라우팅을 소유합니다.
+꺼져 있으면 일반 `oauthAccountFailover` 정책이 여전히 긴급 429 failover를 제공할 수 있습니다.
+
 | 키 | 타입 | 기본값 | 설명 |
 | --- | --- | --- | --- |
 | `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | 프로세스 로컬 세션 결속과 최종 429 또는 노출된 402 응답에 대한 제한된 failover를 켭니다. |
diff --git a/docs-site/src/content/docs/ko/reference/management-api.md b/docs-site/src/content/docs/ko/reference/management-api.md
index 8aa5aa32a2..825b47485a 100644
--- a/docs-site/src/content/docs/ko/reference/management-api.md
+++ b/docs-site/src/content/docs/ko/reference/management-api.md
@@ -169,14 +169,21 @@ Authorization: Bearer <admin-token>
 `enabled`(boolean), `autoSwitchThreshold`(0~100 정수), `strategy`(`quota`, `round-robin`,
 `fill-first`), `stickyLimit`(1~100 정수)을 받을 수 있습니다. 생략된 필드는 현재값 또는 기본값을
 유지합니다. 성공 응답은 최상위에 `{ ok: true, provider, enabled, autoSwitchThreshold, strategy,
-stickyLimit, experimental: true }`를 반환하며 `config` 래퍼는 없습니다. 잘못된 형식이나 객체가 아닌
+`stickyLimit, experimental: true }`를 반환하며 `config` 래퍼는 없습니다. 잘못된 형식이나 객체가 아닌
 본문, 지원되지 않는 provider, 잘못된 필드는 일반 HTTP 400 오류 응답을 반환합니다.
 
+`google-antigravity`의 경우 `enabled: false`를 쓰면
+`providers.google-antigravity.oauthAccountFailover.enabled: false`도 저장됩니다. 이렇게 하면 특수
+할당량 인식 풀과, 그렇지 않으면 계정 존재 여부에 따라 활성화되는 일반 429 fallback을 모두
+비활성화하여 관리 API/CLI에서 “off”가 실제로 엄격한 단일 계정 모드를 뜻하게 됩니다.
+`googleAntigravityAccountPool.enabled`만 직접 편집하면 일반 정책은 변경되지 않습니다.
+
 `POST /api/oauth/accounts/clear-cooldown`의 본문은 `{ provider, accountId }`입니다. provider는
 `anthropic` 또는 `google-antigravity`여야 하며 account는 해당 provider 안에 존재해야 합니다. 알 수
 없는 account는 일반 HTTP 400 `account not found` 오류를 반환합니다. 알려진 account가 cooldown 상태가
-아니어도 HTTP 200과 `{ ok: true, cleared: false }`를 반환하는 멱등 성공입니다. 이 응답에는 token,
-이메일 주소 또는 전체 key가 포함되지 않습니다.
+아니어도 HTTP 200과 `{ ok: true, cleared: false }`를 반환하는 멱등 성공입니다. Google의 경우 이 작업은
+특수 풀 상태와 일반 fallback 상태를 모두 지우므로 pool mode가 변경된 뒤에도 유효합니다. 이 응답에는
+token, 이메일 주소 또는 전체 key가 포함되지 않습니다.
 
 자격 증명 목록 응답은 의도적으로 마스킹됩니다. OAuth access token과 완전한 provider API key는 대시보드 클라이언트에 반환되지 않습니다.
 
diff --git a/docs-site/src/content/docs/reference/cli/providers-accounts.md b/docs-site/src/content/docs/reference/cli/providers-accounts.md
index 4a0235d02d..80f1172260 100644
--- a/docs-site/src/content/docs/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/reference/cli/providers-accounts.md
@@ -226,6 +226,11 @@ For Codex, `on` sets 80% and `off` sets 0%. For OAuth pools, `on`/`off` toggle t
 its threshold; `threshold <n>` stores the integer from 0 through 100 and enables the pool when nonzero.
 `status` reads the current policy. Other providers and invalid values exit 1. `--json` returns:
 
+For `google-antigravity`, `off` disables the specialized quota-aware pool and also stores the
+per-provider generic OAuth 429 opt-out, so the command really enters strict single-account mode.
+Directly setting only `googleAntigravityAccountPool.enabled: false` does not change that separate
+generic policy.
+
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
diff --git a/docs-site/src/content/docs/reference/configuration/providers.md b/docs-site/src/content/docs/reference/configuration/providers.md
index 144e312652..9beb09b81c 100644
--- a/docs-site/src/content/docs/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/reference/configuration/providers.md
@@ -86,6 +86,7 @@ differing backup and rewrites known legacy namespaced selected ids to bare ids.
 | `modelContextWindows?` | `Record<string, number>` | Per-model context fallbacks/caps. These override `contextWindow`: an unknown window uses the configured value, while smaller live metadata remains authoritative. |
 | `modelInputModalities?` | `Record<string, string[]>` | Per-model input hints such as `["text"]` or `["text", "image"]`. |
 | `modelMaxInputTokens?` | `Record<string, number>` | Positive per-model max input limits used for catalog auto-compaction hints. |
+| `modelAutoCompactTokenLimits?` | `Record<string, number>` | Positive safe-integer per-model soft auto-compaction budgets. Values can only lower the effective 90%-of-context/max-input envelope and are omitted when no authoritative context window is known. For canonical `openai`, keys must be exact supported native model IDs without provider or account-selector prefixes. Provider PATCH merges entries; set a key to `null` to delete it or the whole field to `null` to clear the map. These `null` tombstones are PATCH-only. |
 | `defaultMaxOutputTokens?` | `number` | Provider-wide `openai-chat` fallback when the client omits `max_output_tokens`. |
 | `modelMaxOutputTokens?` | `Record<string, number>` | Positive per-model `openai-chat` fallback budgets; exact/pattern matches beat the provider default. |
 | `modelCosts?` | `Record<string, Cost4>` | Per-model display prices (USD per 1M tokens), keyed by that provider's exact upstream model id — not a provider identifier or a routed `provider/model` label, e.g. `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`. Any model id is a valid key — custom providers may target any OpenAI-compatible endpoint through the `openai-chat` adapter, and local or internal provider ids work even when they are absent from the built-in catalogs. User-configured prices win over the built-in catalogs in the Logs `~$` and Usage estimates; historical entries are repriced from the current overlay, so editing a price can move past totals. The fallback order is user `modelCosts` → exact official correction → jawcode catalog → expected-price overlay → model-level vendor fallback, and an all-zero entry falls through to the next source in that sequence. Each rate must be a non-negative finite number at most 1,000,000 (USD per 1M tokens); out-of-range rows are rejected by the management boundary and dropped on load. Display-time estimation only: overlays never affect routing, account selection, quotas, or billing. |
@@ -300,6 +301,11 @@ This opt-in pools Google Antigravity OAuth accounts for the `google-antigravity`
 provider only. It is disabled by default. The pool never supplies credentials to Google AI Studio,
 Vertex AI, another provider entry, or an API-key route.
 
+Disabling this specialized pool turns off quota-aware selection, session affinity, and its 402/429
+retry budget. It does not disable the separate presence-driven generic OAuth 429 fallback. To keep
+Google strictly on one account, also set
+`providers.google-antigravity.oauthAccountFailover.enabled` to `false`.
+
 | Key | Type | Default | Description |
 | --- | --- | --- | --- |
 | `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | Enable process-local session affinity and bounded failover on a final 429 or surfaced 402 response. |
@@ -329,8 +335,10 @@ risk.
 ### `oauthAccountFailover`
 
 Rotates to another logged-in account of the same provider when one is rate-limited, for OAuth
-providers that have no pool of their own — xAI, Cursor, Kimi, GitHub Copilot, Google Antigravity,
-and Nous.
+providers while no provider-owned pool is active — xAI, Cursor, Kimi, GitHub Copilot, Google
+Antigravity, and Nous. When `googleAntigravityAccountPool.enabled` is `true`, that specialized pool
+owns Google routing instead; when it is off, this generic policy may still provide emergency 429
+failover.
 
 **Logging in a second account is what turns this on.** With no configuration, rotation activates
 for any of those providers holding 2 or more accounts that are not flagged for reauthentication —
diff --git a/docs-site/src/content/docs/reference/management-api.md b/docs-site/src/content/docs/reference/management-api.md
index 7fa6d9a1ff..62f009f5b8 100644
--- a/docs-site/src/content/docs/reference/management-api.md
+++ b/docs-site/src/content/docs/reference/management-api.md
@@ -196,11 +196,18 @@ values. Success returns top-level `{ ok: true, provider, enabled, autoSwitchThre
 stickyLimit, experimental: true }`; there is no `config` wrapper. A malformed or non-object body,
 an unsupported provider, or any invalid field returns a generic HTTP 400 error response.
 
+For `google-antigravity`, writing `enabled: false` also stores
+`providers.google-antigravity.oauthAccountFailover.enabled: false`. This keeps the management/CLI
+meaning of “off” honest by disabling both the specialized quota-aware pool and the otherwise
+presence-driven generic 429 fallback. Editing only `googleAntigravityAccountPool.enabled` directly
+does not alter the generic policy.
+
 `POST /api/oauth/accounts/clear-cooldown` takes `{ provider, accountId }`. The provider must be
 `anthropic` or `google-antigravity`, and the account must exist within that provider. An unknown
 account returns a generic HTTP 400 `account not found` error. A known account that is not cooled down
-is an idempotent success: HTTP 200 with `{ ok: true, cleared: false }`. These responses never include
-tokens, email addresses, or complete keys.
+is an idempotent success: HTTP 200 with `{ ok: true, cleared: false }`. For Google, the operation
+clears both the specialized pool state and the generic fallback state, so it remains valid after a
+pool-mode change. These responses never include tokens, email addresses, or complete keys.
 
 Credential list responses are deliberately masked. OAuth access tokens and complete provider API
 keys are not returned to dashboard clients.
diff --git a/docs-site/src/content/docs/ru/reference/cli/providers-accounts.md b/docs-site/src/content/docs/ru/reference/cli/providers-accounts.md
index 5ee51191e1..9f26c029d9 100644
--- a/docs-site/src/content/docs/ru/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/ru/reference/cli/providers-accounts.md
@@ -201,6 +201,11 @@ quota-bar'ов дашборда.
 включает пул, если значение ненулевое. `status` читает текущую политику. Для неподдерживаемых
 провайдеров и некорректных значений команда завершается кодом 1. `--json` возвращает:
 
+Для `google-antigravity` команда `off` отключает специализированный пул с учётом квоты и также
+сохраняет per-provider opt-out для generic OAuth-failover при 429, поэтому команда действительно
+переводит провайдера в строгий режим одного аккаунта. Если напрямую задать только
+`googleAntigravityAccountPool.enabled: false`, эта отдельная generic policy не изменится.
+
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
diff --git a/docs-site/src/content/docs/ru/reference/configuration/providers.md b/docs-site/src/content/docs/ru/reference/configuration/providers.md
index 6c267de567..27d552be06 100644
--- a/docs-site/src/content/docs/ru/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ru/reference/configuration/providers.md
@@ -85,6 +85,7 @@ cross-route credential fallback не существует. Строки API GPT-
 | `modelContextWindows?` | `Record<string, number>` | Значения и cap'ы контекста по отдельным моделям. Перекрывают `contextWindow`: если окно неизвестно, берётся заданное значение, а более маленькая live-metadata остаётся авторитетной. |
 | `modelInputModalities?` | `Record<string, string[]>` | Подсказки modality по модели, например `["text"]` или `["text", "image"]`. |
 | `modelMaxInputTokens?` | `Record<string, number>` | Положительные лимиты max input по моделям, используемые для подсказок auto-compaction в каталоге. |
+| `modelAutoCompactTokenLimits?` | `Record<string, number>` | Мягкие бюджеты автосжатия по моделям в виде положительных безопасных целых чисел. Они могут только уменьшать эффективную границу в 90 % контекста или максимального ввода и не выдаются, если авторитетное окно контекста неизвестно. Для канонического `openai` ключами могут быть только точные поддерживаемые ID нативных моделей без префиксов провайдера или селектора аккаунта. PATCH провайдера объединяет записи: `null` для ключа удаляет его, а `null` для всего поля очищает карту. Такие маркеры `null` допустимы только в PATCH. |
 | `defaultMaxOutputTokens?` | `number` | Provider-wide fallback для `openai-chat`, когда клиент не передал `max_output_tokens`. |
 | `modelMaxOutputTokens?` | `Record<string, number>` | Положительные fallback-budget'ы `openai-chat` по моделям; exact/pattern-match имеет приоритет над provider-default. |
 | `modelCosts?` | `Record<string, Cost4>` | Отображаемые цены по моделям (USD за 1M токенов), ключ — точный upstream id модели этого провайдера (не идентификатор провайдера и не маршрутизируемая метка `provider/model`), значение — четыре поля: `input`, `output`, `cacheRead`, `cacheWrite` (пример: `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`). Любой id допустим — кастомный провайдер может указывать на любой OpenAI-совместимый endpoint через адаптер `openai-chat`, а локальные и внутренние провайдеры работают даже без строки во встроенных каталогах. Пользовательские цены имеют приоритет над встроенными каталогами в оценках `~$` в Logs и Usage; исторические записи пересчитываются по текущему оверлею, поэтому изменение цены может сдвинуть прошлые суммы (порядок: пользователь → каталог jawcode → expected-price overlay → вендорская цена модели); полностью нулевая запись переходит к следующему источнику. Каждая ставка должна быть неотрицательным конечным числом не более 1 000 000 (USD за 1M токенов); строки вне диапазона отклоняются на управляющей границе и отбрасываются при загрузке. Только оценка для отображения: оверлеи не влияют на маршрутизацию, выбор аккаунта, квоты или биллинг. |
@@ -215,6 +216,15 @@ backoff и может переключить аккаунт уже внутри
 провайдера `google-antigravity`. По умолчанию она выключена. Пул никогда не передаёт credentials в
 Google AI Studio, Vertex AI, другую запись провайдера или маршрут с API-ключом.
 
+Отключение этого специализированного пула выключает выбор с учётом квоты, привязку сессий и его
+бюджет повторов для 402/429. Оно не отключает отдельный generic OAuth-failover при 429, который
+активируется наличием аккаунтов. Чтобы Google строго использовал один аккаунт, также задайте
+`providers.google-antigravity.oauthAccountFailover.enabled` равным `false`.
+
+Когда `googleAntigravityAccountPool.enabled` равен `true`, маршрутизацией Google владеет этот
+специализированный пул. Когда он выключен, отдельная generic policy всё ещё может выполнять
+аварийный failover при 429.
+
 | Ключ | Тип | По умолчанию | Описание |
 | --- | --- | --- | --- |
 | `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | Включить локальную для процесса привязку сессий и ограниченный failover после финального 429 или дошедшего до прокси 402. |
diff --git a/docs-site/src/content/docs/ru/reference/management-api.md b/docs-site/src/content/docs/ru/reference/management-api.md
index 0db883fba7..caa0c68fe8 100644
--- a/docs-site/src/content/docs/ru/reference/management-api.md
+++ b/docs-site/src/content/docs/ru/reference/management-api.md
@@ -191,11 +191,18 @@ autoSwitchThreshold, strategy, stickyLimit, experimental: true }`; обёртк
 или не являющийся объектом body, неподдерживаемый provider или некорректное поле возвращает обычный
 ответ с ошибкой HTTP 400.
 
+Для `google-antigravity` запись `enabled: false` также сохраняет
+`providers.google-antigravity.oauthAccountFailover.enabled: false`. Благодаря этому значение «off»
+в management API/CLI действительно отключает как специализированный пул с учётом квоты, так и
+generic fallback при 429, который в противном случае активируется наличием аккаунтов. Прямое
+редактирование только `googleAntigravityAccountPool.enabled` не изменяет эту generic policy.
+
 `POST /api/oauth/accounts/clear-cooldown` принимает `{ provider, accountId }`. Provider должен быть
 `anthropic` или `google-antigravity`, а аккаунт должен существовать внутри этого provider. Неизвестный
 аккаунт возвращает обычную ошибку HTTP 400 `account not found`. Известный аккаунт без cooldown даёт
-идемпотентный успех: HTTP 200 с `{ ok: true, cleared: false }`. Эти ответы не содержат token'ов,
-email-адресов или полных key.
+идемпотентный успех: HTTP 200 с `{ ok: true, cleared: false }`. Для Google операция очищает как
+состояние специализированного пула, так и состояние generic fallback, поэтому остаётся допустимой
+после смены pool-mode. Эти ответы не содержат token'ов, email-адресов или полных key.
 
 Ответы со списками credential'ов намеренно маскируются. OAuth access-token'ы и полные API-key'и
 провайдеров клиентам дашборда не возвращаются.
diff --git a/docs-site/src/content/docs/tr/reference/cli/providers-accounts.md b/docs-site/src/content/docs/tr/reference/cli/providers-accounts.md
index b20f062b3e..9fbbed6475 100644
--- a/docs-site/src/content/docs/tr/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/tr/reference/cli/providers-accounts.md
@@ -242,6 +242,11 @@ havuzlarında `on` / `off`, mevcut eşiği koruyarak havuzun etkin durumunu değ
 havuzu etkinleştirir. `status` geçerli politikayı okur. Desteklenmeyen sağlayıcılar ve
 geçersiz değerler 1 ile çıkar. `--json` şunu döndürür:
 
+`google-antigravity` için `off`, özelleşmiş kota duyarlı havuzu devre dışı bırakır ve ayrıca
+sağlayıcı başına genel OAuth 429 devre dışı bırakma ayarını saklar; böylece komut gerçekten katı tek
+hesap moduna geçer. Yalnızca `googleAntigravityAccountPool.enabled: false` değerini doğrudan ayarlamak
+bu ayrı genel politikayı değiştirmez.
+
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
diff --git a/docs-site/src/content/docs/tr/reference/configuration/providers.md b/docs-site/src/content/docs/tr/reference/configuration/providers.md
index 7f0ae6628f..6c2110b7af 100644
--- a/docs-site/src/content/docs/tr/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/tr/reference/configuration/providers.md
@@ -91,6 +91,7 @@ alanlı seçilmiş kimlikleri yalın kimliklere yeniden yazar.
 | `modelContextWindows?` | `Record<string, number>` | Model başına bağlam geri dönüşleri/sınırları. Bunlar `contextWindow`'u geçersiz kılar: bilinmeyen bir pencere yapılandırılmış değeri kullanırken, daha küçük canlı meta veriler yetkili kalır. |
 | `modelInputModalities?` | `Record<string, string[]>` | Model başına girdi ipuçları, örn. `["text"]` veya `["text", "image"]`. |
 | `modelMaxInputTokens?` | `Record<string, number>` | Katalog otomatik sıkıştırma ipuçları için kullanılan pozitif model başına maksimum girdi sınırları. |
+| `modelAutoCompactTokenLimits?` | `Record<string, number>` | Model başına pozitif güvenli tamsayı biçiminde yumuşak otomatik sıkıştırma bütçeleri. Değerler yalnızca bağlamın veya maksimum girdinin etkin %90 zarfını düşürebilir ve yetkili bir bağlam penceresi bilinmiyorsa yayımlanmaz. Canonical `openai` için anahtarlar, sağlayıcı veya hesap seçici öneki olmadan desteklenen tam yerel model kimlikleri olmalıdır. Sağlayıcı PATCH girdileri birleştirir; bir anahtarı `null` yapmak o anahtarı siler, alanın tamamını `null` yapmak haritayı temizler. Bu `null` silme işaretleri yalnızca PATCH içindir. |
 | `defaultMaxOutputTokens?` | `number` | İstemci `max_output_tokens` değerini atladığında sağlayıcı genelinde `openai-chat` geri dönüşü. |
 | `modelMaxOutputTokens?` | `Record<string, number>` | Pozitif model başına `openai-chat` geri dönüş bütçeleri; tam/kalıp eşleşmeleri sağlayıcı varsayılanını yener. |
 | `modelCosts?` | `Record<string, Cost4>` | Sağlayıcının tam yukarı akış model kimliğine göre anahtarlanan model başına görüntüleme fiyatları (1M token başına USD) — bir sağlayıcı tanımlayıcısı veya yönlendirilen `provider/model` etiketi değil, örn. `{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`. Herhangi bir model kimliği geçerli bir anahtardır — özel sağlayıcılar `openai-chat` adaptörü aracılığıyla herhangi bir OpenAI uyumlu uç noktayı hedefleyebilir ve yerel veya dahili sağlayıcı kimlikleri yerleşik kataloglarda bulunmasalar bile çalışır. Kullanıcı tarafından yapılandırılan fiyatlar Günlükler `~$` ve Kullanım tahminlerinde yerleşik katalogları yener; geçmiş girdiler geçerli katmandan yeniden fiyatlandırılır, bu nedenle bir fiyatı düzenlemek geçmiş toplamları değiştirebilir. Geri dönüş sırası: kullanıcı `modelCosts` → jawcode kataloğu → beklenen fiyat katmanı → model düzeyinde satıcı geri dönüşü ve tamamen sıfır bir girdi bu dizideki bir sonraki kaynağa düşer. Her oran en fazla 1.000.000 (1M token başına USD) olan negatif olmayan sonlu bir sayı olmalıdır; aralık dışı satırlar yönetim sınırı tarafından reddedilir ve yükleme sırasında bırakılır. Yalnızca görüntüleme zamanı tahmini: katmanlar yönlendirmeyi, hesap seçimini, kotaları veya faturalandırmayı asla etkilemez. |
@@ -239,6 +240,11 @@ Bu isteğe bağlı özellik Google Antigravity OAuth hesaplarını yalnızca `go
 Assist sağlayıcısı için havuzlar. Varsayılan olarak devre dışıdır. Havuz, kimlik bilgilerini hiçbir
 zaman Google AI Studio, Vertex AI, başka bir sağlayıcı kaydı veya API anahtarı rotasına vermez.
 
+Bu özelleşmiş havuzu devre dışı bırakmak kota duyarlı seçimi, oturum bağlılığını ve kendi 402/429
+yeniden deneme bütçesini kapatır. Ayrı, varlığa dayalı genel OAuth 429 yük devretmesini devre dışı
+bırakmaz. Google'ı katı biçimde tek hesapta tutmak için ayrıca
+`providers.google-antigravity.oauthAccountFailover.enabled` değerini `false` olarak ayarlayın.
+
 | Anahtar | Tip | Varsayılan | Açıklama |
 | --- | --- | --- | --- |
 | `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | İşleme özgü oturum bağlılığını ve son 429 ya da yüzeye çıkan 402 yanıtında sınırlı yük devretmeyi etkinleştirir. |
@@ -265,6 +271,66 @@ kullanmayın. Çoklu hesap otomasyonu sağlayıcı koşullarını ihlal edebilir
 özelliği devre dışı bırakın.
 :::
 
+### `oauthAccountFailover`
+
+Bir hesap hız sınırına takıldığında, sağlayıcıya ait bir havuz etkin değilken aynı OAuth
+sağlayıcısının oturum açılmış başka bir hesabına geçer — xAI, Cursor, Kimi, GitHub Copilot, Google
+Antigravity ve Nous. `googleAntigravityAccountPool.enabled` değeri `true` olduğunda Google
+yönlendirmesini bunun yerine bu özelleşmiş havuz yönetir; kapalı olduğunda bu genel politika acil 429
+yük devretmesi sağlamaya devam edebilir.
+
+**İkinci bir hesapla oturum açmak bunu etkinleştirir.** Yapılandırma olmadan, bu sağlayıcılardan yeniden
+kimlik doğrulama için işaretlenmemiş 2 veya daha fazla hesaba sahip herhangi biri için rotasyon
+etkinleşir — `apiKeyPool`'un 2+ anahtarlı bir havuza zaten uyguladığı kuralla aynıdır. Saklanan tek
+hesabı olan bir sağlayıcı tam olarak önceki gibi davranır.
+
+| Anahtar | Tip | Varsayılan | Açıklama |
+| --- | --- | --- | --- |
+| `oauthAccountFailover.enabled?` | `boolean` | varlığa dayalı | Genel geçersiz kılma. `false` her yerde tek hesap davranışını zorlar; `true` rotasyonu zorla etkinleştirir. |
+| `providers.<name>.oauthAccountFailover.enabled?` | `boolean` | devralır | Sağlayıcı başına geçersiz kılma; genel ayarı ve hesap varlığını geçersiz kılar. |
+
+Koşullarını sınamak istemediğiniz bir sağlayıcıda katı tek hesap davranışını korumak için:
+
+```json
+{
+  "providers": {
+    "cursor": {
+      "oauthAccountFailover": { "enabled": false }
+    }
+  }
+}
+```
+
+Bu ayar oturum açma, hesap ekleme ve yeniden kimlik doğrulama işlemleri boyunca korunur.
+
+`anthropicAccountPool`'dan kasıtlı olarak daha dardır: oturum bağlılığı, kota sıralamalı seçim ve prob
+kiralamaları yoktur. Yalnızca tek bir soruyu yanıtlar — az önce 429 döndüren hesap soğumaya alındı;
+kullanılabilir başka bir hesap var mı?
+
+Codex havuzu ve Anthropic havuzu kapsam dışıdır ve kendi rotasyonlarını sürdürür; bunu etkinleştirmek
+ikisini de değiştirmez. Saklanan tek hesabı olan bir sağlayıcı için kesinlikle işlem yapılmaz ve hiçbir
+soğuma kaydedilmez.
+
+429 durumunda başarısız hesap, mevcut olduğunda `Retry-After` kullanılarak (en fazla 15 dakika) veya
+varsayılan bir geri çekilmeyle soğumaya alınır ve istek, istek başına en fazla üç rotasyonla sonraki
+uygun hesapta yeniden oynatılır. Yeniden kimlik doğrulama için işaretlenen bir hesap asla seçilmez.
+Soğuma süreleri işleme özgüdür; yeniden başlatma bunları unutur.
+
+Rotasyon yalnızca bearer belirtecini değil, alternatif hesabın **tam** kimlik bilgisi anlık görüntüsünü
+taşır. Böylece yönlendirme meta verilerini belirteciyle eşleştiren bir sağlayıcı — örneğin
+Antigravity'nin Cloud Code Assist proje kimliği — bir hesabın belirtecini başka bir hesabın meta
+verileriyle gönderemez.
+
+Geçerli kapsam sıradan Responses istek yollarıdır. Cursor, hız sınırlarını HTTP durumu yerine adaptör
+olayları olarak bildirir ve bağımsız Antigravity görüntü uç noktasının kendi istek yolu vardır; henüz
+ikisi de rotasyon yapmaz.
+
+:::caution[Deneysel]
+Abonelik hesapları arasında rotasyon yapmak ikinci bir hesabın kotasını harcar ve bazı sağlayıcıların
+koşullarını ihlal edebilir. İstediğiniz ödünleşim bu değilse genel olarak veya söz konusu sağlayıcı için
+`enabled: false` ayarlayın.
+:::
+
 ### Yönetilen kayıt biçimleri
 
 `apiKeys[]` girdileri `id`, `name`, oluşturulan `key` ve ISO `createdAt`
diff --git a/docs-site/src/content/docs/tr/reference/management-api.md b/docs-site/src/content/docs/tr/reference/management-api.md
index a53f2c451a..8f24b8b8b2 100644
--- a/docs-site/src/content/docs/tr/reference/management-api.md
+++ b/docs-site/src/content/docs/tr/reference/management-api.md
@@ -208,11 +208,18 @@ autoSwitchThreshold, strategy, stickyLimit, experimental: true }` döndürür; `
 yoktur. Bozuk veya nesne olmayan gövde, desteklenmeyen provider ya da geçersiz alan genel bir HTTP
 400 hata yanıtı döndürür.
 
+`google-antigravity` için `enabled: false` yazmak ayrıca
+`providers.google-antigravity.oauthAccountFailover.enabled: false` değerini saklar. Bu, hem özelleşmiş
+kota duyarlı havuzu hem de aksi halde varlığa dayalı genel 429 yük devretmesini devre dışı bırakarak
+yönetim/CLI'daki `off` anlamını dürüst kılar. Yalnızca `googleAntigravityAccountPool.enabled` değerini
+doğrudan düzenlemek genel politikayı değiştirmez.
+
 `POST /api/oauth/accounts/clear-cooldown`, `{ provider, accountId }` gövdesini alır. Provider
 `anthropic` veya `google-antigravity` olmalı ve account ilgili provider içinde bulunmalıdır. Bilinmeyen
 account, genel HTTP 400 `account not found` hatası döndürür. Cooldown durumunda olmayan bilinen bir
-account için işlem idempotent olarak başarılıdır: HTTP 200 ve `{ ok: true, cleared: false }`. Bu
-yanıtlar token, e-posta adresi veya eksiksiz key içermez.
+account için işlem idempotent olarak başarılıdır: HTTP 200 ve `{ ok: true, cleared: false }`. Google
+için işlem hem özelleşmiş havuz durumunu hem de genel yük devretme durumunu temizler; böylece havuz modu
+değişikliğinden sonra da geçerli kalır. Bu yanıtlar token, e-posta adresi veya eksiksiz key içermez.
 
 Kimlik bilgisi listesi yanıtları kasıtlı olarak maskelenir. OAuth erişim
 belirteçleri ve eksiksiz sağlayıcı API anahtarları kontrol paneli istemcilerine
diff --git a/docs-site/src/content/docs/zh-cn/reference/cli/providers-accounts.md b/docs-site/src/content/docs/zh-cn/reference/cli/providers-accounts.md
index e0e4b4e153..24056be4a4 100644
--- a/docs-site/src/content/docs/zh-cn/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/zh-cn/reference/cli/providers-accounts.md
@@ -182,6 +182,10 @@ token，也不是简单重读账号列表。`--json` 返回
 会切换池的启用状态，同时保留现有阈值。`threshold <n>` 会保存 0 到 100 的整数，且非零值
 会启用池。`status` 读取当前策略。不支持的提供方和无效值都会以 1 退出。`--json` 返回：
 
+对于 `google-antigravity`，`off` 会禁用专用的配额感知池，并同时保存按提供方设置的通用 OAuth
+429 停用项，因此该命令会真正进入严格的单账户模式。仅直接设置
+`googleAntigravityAccountPool.enabled: false` 不会改变这项独立的通用策略。
+
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
diff --git a/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md b/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
index 059157efab..a54afdac23 100644
--- a/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
@@ -72,6 +72,7 @@ selector，而不是分配一个新名称。
 | `modelContextWindows?` | `Record<string, number>` | 按模型设置的上下文数值与上限。优先于 `contextWindow`：窗口未知时采用所配置的数值，而更小的实时元数据仍然优先。 |
 | `modelInputModalities?` | `Record<string, string[]>` | 按模型设置的输入提示，例如 `["text"]` 或 `["text", "image"]`。 |
 | `modelMaxInputTokens?` | `Record<string, number>` | 正数型、按模型设置的最大输入限制，用于目录自动压缩提示。 |
+| `modelAutoCompactTokenLimits?` | `Record<string, number>` | 按模型设置的正安全整数软自动压缩预算。该值只能降低“上下文或最大输入的 90%”这一有效上限；没有已知的权威上下文窗口时不会输出。对于规范 `openai`，键必须是受支持的精确原生模型 ID，且不得包含提供者或账户选择器前缀。提供者 PATCH 会合并条目；将某个键设为 `null` 会删除该键，将整个字段设为 `null` 会清空映射。这些 `null` 删除标记仅适用于 PATCH。 |
 | `defaultMaxOutputTokens?` | `number` | 当客户端省略 `max_output_tokens` 时，`openai-chat` 的提供者级回退值。 |
 | `modelMaxOutputTokens?` | `Record<string, number>` | 正数型、按模型设置的 `openai-chat` 回退预算；精确/模式匹配优先于提供者默认值。 |
 | `modelCosts?` | `Record<string, Cost4>` | 按模型设置的显示价格（每 100 万 token 的美元数），以该提供者的精确上游模型 ID 为键（不是提供者标识符或路由后的 `provider/model` 标签），值为四个字段：`input`、`output`、`cacheRead`、`cacheWrite`（示例：`{ "deepseek-v4-flash": { "input": 0.14, "output": 0.28, "cacheRead": 0.0028, "cacheWrite": 0 } }`）。任何模型 ID 都是有效键——自定义提供者可以通过 `openai-chat` 适配器指向任意 OpenAI 兼容端点，即使不存在于内置目录中，本地 OpenAI 兼容和内部提供者的 ID 同样有效。用户配置的价格在 Logs 的 `~$` 和 Usage 估算中优先于内置目录；历史条目也会按当前覆盖项重新计价，因此修改价格可能改变过去的总额（回退顺序：用户配置 → jawcode 目录 → expected-price 覆盖 → 模型级厂商价格）；全零条目会回退到该顺序中的下一个来源。每个费率必须是大于等于 0 的有限数字，且不超过 1,000,000（每 100 万 token 的美元数）；超出范围的条目会在管理边界被拒绝，并在加载时被丢弃。仅用于显示的估算：覆盖项不影响路由、账户选择、配额或计费。 |
@@ -171,6 +172,10 @@ affinity。这些策略不能规避 provider enforcement。
 此可选功能仅为 `google-antigravity` Cloud Code Assist 提供方池化 Google Antigravity OAuth
 账户，默认关闭。该池绝不会把凭据用于 Google AI Studio、Vertex AI、其他提供方条目或 API 密钥路由。
 
+禁用此专用池会关闭配额感知选择、会话亲和性及其 402/429 重试预算，但不会禁用另一项由账户
+存在情况驱动的通用 OAuth 429 故障转移。要让 Google 严格只使用一个账户，还需将
+`providers.google-antigravity.oauthAccountFailover.enabled` 设为 `false`。
+
 | 键 | 类型 | 默认值 | 说明 |
 | --- | --- | --- | --- |
 | `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | 启用进程本地会话亲和性，以及最终 429 或透传到此处的 402 响应触发的有界故障转移。 |
@@ -194,6 +199,58 @@ Assist `projectId`。最终 429 或透传到此处的 402 会让失败账户进
 条款；如果不接受这一风险，请保持关闭。
 :::
 
+### `oauthAccountFailover`
+
+当没有提供方自有的池处于活动状态时，如果某个账户受到限流，此策略会为 OAuth 提供方轮换到
+同一提供方的另一个已登录账户——适用于 xAI、Cursor、Kimi、GitHub Copilot、Google Antigravity
+和 Nous。当 `googleAntigravityAccountPool.enabled` 为 `true` 时，该专用池会接管 Google 路由；
+关闭时，这项通用策略仍可提供紧急 429 故障转移。
+
+**登录第二个账户就是开启此功能的方式。** 未配置时，只要上述任一提供方持有 2 个或更多未标记为
+需要重新认证的账户，轮换就会启用——与 `apiKeyPool` 已对包含 2 个以上密钥的密钥池采用的规则
+相同。只存储一个账户的提供方仍保持原有行为。
+
+| 键 | 类型 | 默认值 | 说明 |
+| --- | --- | --- | --- |
+| `oauthAccountFailover.enabled?` | `boolean` | 由账户存在情况驱动 | 全局覆盖项。`false` 强制所有提供方保持单账户行为；`true` 强制启用轮换。 |
+| `providers.<name>.oauthAccountFailover.enabled?` | `boolean` | 继承 | 按提供方设置的覆盖项；优先级高于全局设置和账户存在情况。 |
+
+若要让某个不想试探其条款的提供方严格保持单账户行为：
+
+```json
+{
+  "providers": {
+    "cursor": {
+      "oauthAccountFailover": { "enabled": false }
+    }
+  }
+}
+```
+
+该设置在登录、添加账户和重新认证后都会保留。
+
+此功能刻意比 `anthropicAccountPool` 更窄：没有会话亲和性、没有按配额排名的选择，也没有探测租约。
+它只回答一个问题——刚刚返回 429 的账户已进入冷却，是否还有另一个账户可用。
+
+Codex 池和 Anthropic 池不在此功能范围内，并保留各自的轮换；启用此功能不会改变两者。仅存储一个
+账户的提供方严格不执行任何操作，也不会为其记录冷却。
+
+收到 429 时，失败账户会按 `Retry-After`（上限 15 分钟）或默认退避进入冷却，并在下一个合格账户
+上重放请求；每个请求最多轮换三次。标记为需要重新认证的账户永远不会被选中。冷却仅存在于进程
+本地，因此重启后会被遗忘。
+
+轮换会携带替代账户的**完整**凭据快照，而不只是 bearer；这样，对于将路由元数据与 token 配对的
+提供方——例如 Antigravity 的 Cloud Code Assist 项目 id——就不会把一个账户的 token 与另一个
+账户的元数据一起发送。
+
+当前范围是普通 Responses 请求路径。Cursor 会以适配器事件而不是 HTTP 状态报告限流，而独立的
+Antigravity 图像端点拥有自己的请求路径；两者目前都不会轮换。
+
+:::caution[实验性]
+在订阅账户之间轮换会消耗第二个账户的配额，也可能违反某些提供方的条款。如果你不接受这种取舍，
+请在全局或相应提供方上设置 `enabled: false`。
+:::
+
 ### 托管记录形状
 
 `apiKeys[]` 条目包含 `id`、`name`、生成的 `key` 以及 ISO 格式的 `createdAt` 字符串。`codexAccounts[]` 条目要求有 `id`、`email` 和 `isMain`，并可选 `plan`、`chatgptAccountId` 和具备隐私安全性的 `logLabel`。这些记录通常由仪表板管理。
diff --git a/docs-site/src/content/docs/zh-cn/reference/management-api.md b/docs-site/src/content/docs/zh-cn/reference/management-api.md
index 6dc075c9c2..d28df34545 100644
--- a/docs-site/src/content/docs/zh-cn/reference/management-api.md
+++ b/docs-site/src/content/docs/zh-cn/reference/management-api.md
@@ -172,10 +172,16 @@ Authorization: Bearer <admin-token>
 experimental: true }`，没有 `config` 包装层。格式错误或非对象的请求体、不支持的 provider 或
 无效字段都会返回通用 HTTP 400 错误响应。
 
+对于 `google-antigravity`，写入 `enabled: false` 还会保存
+`providers.google-antigravity.oauthAccountFailover.enabled: false`。这使管理 API/CLI 中“关闭”的
+含义名副其实：它会同时禁用专用的配额感知池和原本由账户存在情况驱动的通用 429 回退。仅直接编辑
+`googleAntigravityAccountPool.enabled` 不会改变这项通用策略。
+
 `POST /api/oauth/accounts/clear-cooldown` 的请求体为 `{ provider, accountId }`。provider 必须是
 `anthropic` 或 `google-antigravity`，且账号必须存在于该 provider 内。未知账号返回通用 HTTP 400
 `account not found` 错误。已知账号即使不在冷却中也会幂等成功：HTTP 200 和
-`{ ok: true, cleared: false }`。这些响应不会包含 token、电子邮件地址或完整密钥。
+`{ ok: true, cleared: false }`。对于 Google，此操作会同时清除专用池状态和通用回退状态，因此
+即使池模式发生变化，该操作仍然有效。这些响应不会包含 token、电子邮件地址或完整密钥。
 
 凭证列表响应会刻意脱敏。OAuth 访问令牌和完整的 provider API 密钥不会返回给仪表板客户端。
 
diff --git a/docs-site/src/content/docs/zh-tw/reference/cli/providers-accounts.md b/docs-site/src/content/docs/zh-tw/reference/cli/providers-accounts.md
index d7c1479993..a245fab749 100644
--- a/docs-site/src/content/docs/zh-tw/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/zh-tw/reference/cli/providers-accounts.md
@@ -142,6 +142,10 @@ List and switch provider accounts and API-key pools (masked output only).
 會切換池的啟用狀態，同時保留現有閾值。`threshold <n>` 會儲存 0 到 100 的整數，且非零值
 會啟用池。`status` 讀取目前政策。不支援的供應商與無效值會以 1 離開。`--json` 回傳：
 
+對 `google-antigravity` 而言，`off` 會停用專用的配額感知池，並同時儲存該供應商的通用 OAuth
+429 停用設定，因此此指令確實會進入嚴格的單一帳號模式。只直接設定
+`googleAntigravityAccountPool.enabled: false` 並不會變更那項獨立的通用政策。
+
 ```text
 { provider, autoSwitchThreshold: number, enabled: boolean }
 ```
@@ -174,7 +178,8 @@ List and switch provider accounts and API-key pools (masked output only).
 清除 `openai` Codex 池或支援的 `anthropic`、`google-antigravity` OAuth 池中一個 runtime 冷卻。
 `main` 只適用於 `openai`，並會在 JSON 中正規化為 `__main__`。無效供應商或未知帳號會被 API
 以 400 拒絕，CLI 以 1 離開。對於沒有作用中冷卻的真實帳號，操作會冪等成功：API 回傳 200、
-CLI 以 0 離開，且 `cleared` 為 `false`。`--json` 嚴格回傳以下結構：
+CLI 以 0 離開，且 `cleared` 為 `false`。對 `google-antigravity`，此操作會同時清除專用池狀態
+與通用 fallback 狀態，因此在池模式變更後仍然有效。`--json` 嚴格回傳以下結構：
 
 ```text
 { ok: true, provider, id, cleared: boolean }
diff --git a/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md b/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
index c74f795431..a2d5354352 100644
--- a/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
@@ -54,6 +54,7 @@ description: 供應商項目、認證、端點、模型目錄、配額、context
 | `modelContextWindows?` | `Record<string, number>` | Per-model context 上限。這些覆寫 `contextWindow` 且永不提高較小的即時中繼資料。 |
 | `modelInputModalities?` | `Record<string, string[]>` | Per-model 輸入提示，如 `["text"]` 或 `["text", "image"]`。 |
 | `modelMaxInputTokens?` | `Record<string, number>` | 用於目錄自動壓縮提示的正數 per-model max input 限制。 |
+| `modelAutoCompactTokenLimits?` | `Record<string, number>` | Per-model 正安全整數型 soft 自動壓縮預算。此值只能降低「context 或 max input 的 90%」這個有效上限；沒有已知的權威 context window 時不會輸出。對 canonical `openai` 而言，key 必須是受支援的精確 native model ID，且不得含 provider 或 account-selector 前綴。Provider PATCH 會合併項目；將單一 key 設為 `null` 會刪除該 key，將整個欄位設為 `null` 會清空 map。這些 `null` tombstone 僅供 PATCH 使用。 |
 | `defaultMaxOutputTokens?` | `number` | 當客戶端省略 `max_output_tokens` 時的供應商範圍 `openai-chat` 後備。 |
 | `modelMaxOutputTokens?` | `Record<string, number>` | 正數 per-model `openai-chat` 後援預算；精確／模式比對勝過供應商預設。 |
 | `headers?` | `Record<string, string>` | 額外上游標頭。Authorization、cookie、API-key 標頭、內嵌換行與無效名稱被拒絕。 |
@@ -140,6 +141,10 @@ API-key 供應商可持有字面值金鑰或環境參考。OAuth 供應商使用
 此選用功能只為 `google-antigravity` Cloud Code Assist 供應商池化 Google Antigravity OAuth
 帳號，預設停用。此池絕不會把憑證用於 Google AI Studio、Vertex AI、其他供應商項目或 API-key 路由。
 
+停用此專用池會關閉配額感知選擇、session 親和性及其 402/429 重試額度，但不會停用另一個
+由帳號存在情況驅動的通用 OAuth 429 容錯移轉。若要讓 Google 嚴格只使用單一帳號，還必須將
+`providers.google-antigravity.oauthAccountFailover.enabled` 設為 `false`。
+
 | Key | 型別 | 預設值 | 說明 |
 | --- | --- | --- | --- |
 | `googleAntigravityAccountPool.enabled?` | `boolean` | `false` | 啟用行程本地的 session 親和性，以及最終 429 或傳到此處的 402 回應所觸發的有界容錯移轉。 |
@@ -163,6 +168,13 @@ Assist `projectId`。最終 429 或傳到此處的 402 會讓失敗帳號進入
 條款；若不接受這項風險，請保持停用。
 :::
 
+### `oauthAccountFailover`
+
+當沒有供應商自有的池處於啟用狀態時，若某個 OAuth 帳號遭到限流，此機制會輪換到同一供應商的
+另一個已登入帳號，適用於 xAI、Cursor、Kimi、GitHub Copilot、Google Antigravity 與 Nous。
+當 `googleAntigravityAccountPool.enabled` 為 `true` 時，該專用池會接管 Google 路由；當它關閉時，
+這項通用政策仍可能提供緊急 429 容錯移轉。
+
 ### 受管記錄結構
 
 `apiKeys[]` 項目包含 `id`、`name`、生成的 `key` 與 ISO `createdAt` 字串。
diff --git a/docs-site/src/content/docs/zh-tw/reference/management-api.md b/docs-site/src/content/docs/zh-tw/reference/management-api.md
index 38ac5841e8..a5ba8cd359 100644
--- a/docs-site/src/content/docs/zh-tw/reference/management-api.md
+++ b/docs-site/src/content/docs/zh-tw/reference/management-api.md
@@ -172,10 +172,16 @@ Session 簽發在需要 data-plane 認證時停用，這包含遠端綁定。遠
 stickyLimit, experimental: true }`，沒有 `config` 包裝層。格式錯誤或非物件的 body、不支援的
 provider 或無效欄位都會回傳通用 HTTP 400 錯誤回應。
 
+對 `google-antigravity` 寫入 `enabled: false` 時，也會儲存
+`providers.google-antigravity.oauthAccountFailover.enabled: false`。這可確保管理 API／CLI 中
+「off」的語意一致：專用的配額感知池與原本由帳號存在情況驅動的通用 429 容錯移轉都會停用。
+只直接編輯 `googleAntigravityAccountPool.enabled` 並不會變更該通用政策。
+
 `POST /api/oauth/accounts/clear-cooldown` 的 body 為 `{ provider, accountId }`。provider 必須是
 `anthropic` 或 `google-antigravity`，且帳號必須存在於該 provider 內。未知帳號回傳通用 HTTP 400
 `account not found` 錯誤。已知帳號即使不在冷卻中也會冪等成功：HTTP 200 與
-`{ ok: true, cleared: false }`。這些回應不會包含 token、電子郵件地址或完整金鑰。
+`{ ok: true, cleared: false }`。對 Google 而言，此操作會同時清除專用池狀態與通用 fallback
+狀態，因此在池模式變更後仍然有效。這些回應不會包含 token、電子郵件地址或完整金鑰。
 
 憑證清單回應被刻意遮罩。OAuth access token 與完整的供應商 API 金鑰不回傳給儀表板客戶端。
 
diff --git a/src/oauth/generic-account-failover.ts b/src/oauth/generic-account-failover.ts
index 018c46c987..18beb26eba 100644
--- a/src/oauth/generic-account-failover.ts
+++ b/src/oauth/generic-account-failover.ts
@@ -217,6 +217,12 @@ export function genericFailoverRetryAfterSeconds(providerName: string, now = Dat
   return earliest === null ? null : Math.max(1, Math.ceil((earliest - now) / 1000));
 }
 
+/** Clear one account's generic cooldown without disturbing sibling account health. */
+export function clearGenericFailoverAccountHealth(providerName: string, accountId: string): boolean {
+  presence.delete(providerName);
+  return health.delete(healthKey(providerName, accountId));
+}
+
 /** Test seam and manual-recovery hook. */
 export function clearGenericFailoverHealth(providerName?: string): void {
   if (!providerName) {
diff --git a/src/server/management/oauth-account-routes.ts b/src/server/management/oauth-account-routes.ts
index 4aad499a90..2b8cdabdda 100644
--- a/src/server/management/oauth-account-routes.ts
+++ b/src/server/management/oauth-account-routes.ts
@@ -413,6 +413,15 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
       ...(strategy !== undefined ? { strategy } : {}),
       ...(stickyLimit !== undefined ? { stickyLimit } : {}),
     });
+    if (provider === "google-antigravity" && body.enabled === false) {
+      const providerConfig = config.providers?.[provider];
+      if (providerConfig) {
+        providerConfig.oauthAccountFailover = {
+          ...providerConfig.oauthAccountFailover,
+          enabled: false,
+        };
+      }
+    }
     saveConfigPreservingClaudeCode(config);
     reconcileLiveStateStores();
     return jsonResponse({
@@ -437,9 +446,16 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
     if (!getAccountSet(provider)?.accounts.some(account => account.id === accountId)) {
       return jsonResponse({ error: "account not found" }, 400);
     }
-    const cleared = provider === "anthropic"
-      ? (await import("../../oauth/anthropic-routing")).clearAnthropicAccountCooldown(accountId)
-      : (await import("../../oauth/google-antigravity-routing")).clearGoogleAntigravityAccountCooldown(accountId);
+    let cleared: boolean;
+    if (provider === "anthropic") {
+      cleared = (await import("../../oauth/anthropic-routing")).clearAnthropicAccountCooldown(accountId);
+    } else {
+      const specializedCleared = (await import("../../oauth/google-antigravity-routing"))
+        .clearGoogleAntigravityAccountCooldown(accountId);
+      const genericCleared = (await import("../../oauth/generic-account-failover"))
+        .clearGenericFailoverAccountHealth(provider, accountId);
+      cleared = specializedCleared || genericCleared;
+    }
     return jsonResponse({ ok: true, cleared });
   }
 
@@ -514,6 +530,8 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
     const { removeAccount, getAccountSet } = await import("../../oauth/store");
     if (!(await removeAccount(provider, id))) return jsonResponse({ error: "account not found" }, 404);
     reconcileLiveStateStores();
+    const { clearGenericFailoverAccountHealth } = await import("../../oauth/generic-account-failover");
+    clearGenericFailoverAccountHealth(provider, id);
     if (provider === "anthropic") {
       const { clearAnthropicAccountCooldown, clearAnthropicSessionAffinityForAccount } = await import("../../oauth/anthropic-routing");
       clearAnthropicAccountCooldown(id);
diff --git a/src/server/responses/core.ts b/src/server/responses/core.ts
index d69aba5435..ddd19e3562 100644
--- a/src/server/responses/core.ts
+++ b/src/server/responses/core.ts
@@ -2948,13 +2948,15 @@ async function handleResponsesInner(
           accountId: snapshot.accountId,
           generation: snapshot.generation,
         };
+        if (!applyFailoverSnapshot(snapshot)) {
+          return formatErrorResponse(
+            401,
+            "authentication_error",
+            "No eligible Google Antigravity OAuth account available",
+          );
+        }
         bindGoogleAntigravitySessionAffinity(googleAntigravitySessionKey, selection.accountId);
         promoteGoogleAntigravityActiveAccount(selection.accountId);
-        route.provider = {
-          ...route.provider,
-          apiKey: snapshot.accessToken,
-          project: snapshot.projectId,
-        };
         logCtx.provider = formatGoogleAntigravityProviderForLog("google-antigravity", selection.accountId);
       } else {
         const resolved = await getValidAccessTokenSnapshot(route.providerName);
@@ -5302,6 +5304,7 @@ async function handleResponsesInner(
         try { void upstreamResponse.body?.cancel().catch(() => {}); } catch { /* already consumed/closed */ }
         try {
           const snapshot = await getGoogleAntigravityPoolAccessSnapshot(nextAccountId);
+          if (!applyFailoverSnapshot(snapshot)) break;
           bindGoogleAntigravitySessionAffinity(googleAntigravitySessionKey, nextAccountId);
           googleAntigravityPoolAccountId = nextAccountId;
           googleAntigravityPoolFailovers += 1;
@@ -5309,11 +5312,6 @@ async function handleResponsesInner(
             accountId: snapshot.accountId,
             generation: snapshot.generation,
           };
-          route.provider = {
-            ...route.provider,
-            apiKey: snapshot.accessToken,
-            project: snapshot.projectId,
-          };
           invalidateSameTargetRequest();
           promoteGoogleAntigravityActiveAccount(nextAccountId);
           logCtx.provider = formatGoogleAntigravityProviderForLog("google-antigravity", nextAccountId);
@@ -5774,6 +5772,7 @@ async function handleResponsesInner(
             try { void response.body?.cancel().catch(() => {}); } catch { /* already closed */ }
             try {
               const snapshot = await getGoogleAntigravityPoolAccessSnapshot(nextAccountId);
+              if (!applyFailoverSnapshot(snapshot)) break;
               bindGoogleAntigravitySessionAffinity(googleAntigravitySessionKey, nextAccountId);
               googleAntigravityPoolAccountId = nextAccountId;
               googleAntigravityPoolFailovers += 1;
@@ -5781,11 +5780,6 @@ async function handleResponsesInner(
                 accountId: snapshot.accountId,
                 generation: snapshot.generation,
               };
-              route.provider = {
-                ...route.provider,
-                apiKey: snapshot.accessToken,
-                project: snapshot.projectId,
-              };
               invalidateSameTargetRequest();
               promoteGoogleAntigravityActiveAccount(nextAccountId);
               logCtx.provider = formatGoogleAntigravityProviderForLog("google-antigravity", nextAccountId);
diff --git a/src/types/config.ts b/src/types/config.ts
index 2b9a4d1acd..30bdd33753 100644
--- a/src/types/config.ts
+++ b/src/types/config.ts
@@ -634,9 +634,9 @@ export interface OcxConfig {
    * Generic OAuth multi-account 429 failover (#2568). Presence-driven by default.
    *
    * Rotates to another logged-in account of the SAME provider when one is rate-limited, for
-   * OAuth providers that have no pool of their own — xAI, Cursor, Kimi, GitHub Copilot,
+   * OAuth providers while no provider-owned pool is active — xAI, Cursor, Kimi, GitHub Copilot,
    * Antigravity, Nous. The Codex pool and the Anthropic pool own their own rotation and are
-   * excluded; this setting changes neither.
+   * excluded; an enabled Google Antigravity pool likewise owns that request instead.
    *
    * With the key absent, rotation activates when a provider has 2 or more eligible stored
    * accounts — the same consent rule API-key pools already apply to a 2+ key pool (#2568d). A
diff --git a/tests/account-pool-management-api.test.ts b/tests/account-pool-management-api.test.ts
index 012ccada4d..642db1d6d8 100644
--- a/tests/account-pool-management-api.test.ts
+++ b/tests/account-pool-management-api.test.ts
@@ -18,6 +18,12 @@ import {
   resolveGoogleAntigravityAccountForSession,
   rotateGoogleAntigravityAccountOnQuotaError,
 } from "../src/oauth/google-antigravity-routing";
+import {
+  clearGenericFailoverHealth,
+  genericFailoverRetryAfterSeconds,
+  isGenericOAuthFailoverEnabled,
+  rotateGenericOAuthAccountOn429,
+} from "../src/oauth/generic-account-failover";
 import { getAccountSet, removeAccount, saveCredential } from "../src/oauth/store";
 
 function makeCodexConfig(overrides: Partial<OcxConfig> = {}): OcxConfig {
@@ -172,6 +178,7 @@ describe("Anthropic account pool strategy management API", () => {
   }
 
   beforeEach(() => {
+    clearGenericFailoverHealth();
     previousHome = process.env.OPENCODEX_HOME;
     isolatedCodexHome = installIsolatedCodexHome("ocx-pool-mgmt-codex-");
     testDir = mkdtempSync(join(tmpdir(), "ocx-pool-mgmt-"));
@@ -195,6 +202,7 @@ describe("Anthropic account pool strategy management API", () => {
   });
 
   afterEach(() => {
+    clearGenericFailoverHealth();
     if (previousHome === undefined) delete process.env.OPENCODEX_HOME;
     else process.env.OPENCODEX_HOME = previousHome;
     isolatedCodexHome?.restore();
@@ -437,6 +445,52 @@ describe("Anthropic account pool strategy management API", () => {
     }
   });
 
+  test("turning the Google pool off also opts out of presence-driven generic failover", async () => {
+    const config = baseConfig();
+    expect(isGenericOAuthFailoverEnabled(config, "google-antigravity")).toBe(true);
+    const req = new Request("http://localhost/api/oauth/accounts/pool", {
+      method: "PUT",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ provider: "google-antigravity", enabled: false }),
+    });
+    const response = await handleOauthAccountRoutes({
+      req,
+      url: new URL(req.url),
+      config,
+      deps: {},
+      convergeCodexCatalog: async () => ({ status: "unchanged" }),
+      syncClaudeAgentDefsBestEffort: async () => {},
+    });
+
+    expect(response?.status).toBe(200);
+    expect(config.googleAntigravityAccountPool?.enabled).toBe(false);
+    expect(config.providers["google-antigravity"]?.oauthAccountFailover).toEqual({ enabled: false });
+    expect(isGenericOAuthFailoverEnabled(config, "google-antigravity")).toBe(false);
+  });
+
+  test("updating a disabled Google pool without an enabled field preserves generic failover intent", async () => {
+    const config = baseConfig();
+    config.googleAntigravityAccountPool = { enabled: false };
+    config.providers["google-antigravity"]!.oauthAccountFailover = { enabled: true };
+    const req = new Request("http://localhost/api/oauth/accounts/pool", {
+      method: "PATCH",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ provider: "google-antigravity", strategy: "round-robin" }),
+    });
+    const response = await handleOauthAccountRoutes({
+      req,
+      url: new URL(req.url),
+      config,
+      deps: {},
+      convergeCodexCatalog: async () => ({ status: "unchanged" }),
+      syncClaudeAgentDefsBestEffort: async () => {},
+    });
+
+    expect(response?.status).toBe(200);
+    expect(config.providers["google-antigravity"]?.oauthAccountFailover).toEqual({ enabled: true });
+    expect(isGenericOAuthFailoverEnabled(config, "google-antigravity")).toBe(true);
+  });
+
   test("Google Antigravity partial updates normalize invalid current pool fields before saving", async () => {
     const invalidConfig = baseConfig() as OcxConfig & {
       googleAntigravityAccountPool: Record<string, unknown>;
@@ -538,6 +592,34 @@ describe("Anthropic account pool strategy management API", () => {
     }
   });
 
+  test("Google clear-cooldown clears the generic fallback state when the specialized pool is off", async () => {
+    const config = baseConfig();
+    const set = getAccountSet("google-antigravity")!;
+    const aId = set.accounts.find(account => account.id === "google-a")!.id;
+    const bId = set.accounts.find(account => account.id === "google-b")!.id;
+    expect(rotateGenericOAuthAccountOn429(
+      config,
+      "google-antigravity",
+      aId,
+      "120",
+    )).toBe(bId);
+    expect(genericFailoverRetryAfterSeconds("google-antigravity")).toBeGreaterThan(0);
+
+    const server = startServer(0);
+    try {
+      const clear = await fetch(new URL("/api/oauth/accounts/clear-cooldown", server.url), {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({ provider: "google-antigravity", accountId: aId }),
+      });
+      expect(clear.status).toBe(200);
+      expect(await clear.json()).toEqual({ ok: true, cleared: true });
+      expect(genericFailoverRetryAfterSeconds("google-antigravity")).toBeNull();
+    } finally {
+      await server.stop(true);
+    }
+  });
+
   test("deleting Google account state makes a deterministic re-add immediately eligible", async () => {
     const config = baseConfig();
     config.googleAntigravityAccountPool = { enabled: true, strategy: "round-robin", stickyLimit: 5 };
@@ -552,8 +634,25 @@ describe("Anthropic account pool strategy management API", () => {
       email: "delete@example.test",
       projectId: "delete-project",
     });
+    await saveCredential("google-antigravity", {
+      access: "replacement-access",
+      refresh: "replacement-refresh",
+      expires: Date.now() + 3_600_000,
+      accountId: "replacement-identity",
+      email: "replacement@example.test",
+      projectId: "replacement-project",
+    });
     const accountId = getAccountSet("google-antigravity")!.accounts
       .find(account => account.credential.accountId === "delete-identity")!.id;
+    const replacementId = getAccountSet("google-antigravity")!.accounts
+      .find(account => account.credential.accountId === "replacement-identity")!.id;
+    expect(rotateGenericOAuthAccountOn429(
+      config,
+      "google-antigravity",
+      accountId,
+      "120",
+    )).toBe(replacementId);
+    expect(genericFailoverRetryAfterSeconds("google-antigravity")).toBeGreaterThan(0);
     rotateGoogleAntigravityAccountOnQuotaError(config, accountId, "120", "gemini-3.7-flash");
     resetGoogleAntigravityRoutingForManualSelection(accountId);
     bindGoogleAntigravitySessionAffinity("deleted-account-thread", accountId);
@@ -569,17 +668,8 @@ describe("Anthropic account pool strategy management API", () => {
       expect(removed.status).toBe(200);
       expect(getGoogleAntigravityAccountHealthSnapshot(accountId)).toBeNull();
       expect(googleAntigravitySessionAffinitySizeForTests()).toBe(0);
+      expect(genericFailoverRetryAfterSeconds("google-antigravity")).toBeNull();
 
-      await saveCredential("google-antigravity", {
-        access: "replacement-access",
-        refresh: "replacement-refresh",
-        expires: Date.now() + 3_600_000,
-        accountId: "replacement-identity",
-        email: "replacement@example.test",
-        projectId: "replacement-project",
-      });
-      const replacementId = getAccountSet("google-antigravity")!.accounts
-        .find(account => account.credential.accountId === "replacement-identity")!.id;
       await saveCredential("google-antigravity", {
         access: "delete-access-new",
         refresh: "delete-refresh-new",
@@ -591,6 +681,7 @@ describe("Anthropic account pool strategy management API", () => {
       const readdedId = getAccountSet("google-antigravity")!.accounts
         .find(account => account.credential.accountId === "delete-identity")!.id;
       expect(readdedId).toBe(accountId);
+      expect(genericFailoverRetryAfterSeconds("google-antigravity")).toBeNull();
       expect(getEligibleGoogleAntigravityAccounts()).toContain(accountId);
       expect(resolveGoogleAntigravityAccountForSession(
         "post-delete-rotation", "gemini-3.7-flash", config,
diff --git a/tests/config.test.ts b/tests/config.test.ts
index c2626f87bc..7c6bc37974 100644
--- a/tests/config.test.ts
+++ b/tests/config.test.ts
@@ -1692,6 +1692,40 @@ describe("opencodex config defaults", () => {
     expect(readConfigDiagnostics().error).toContain("providers.custom.modelMaxInputTokens");
   });
 
+  test("disk config validates per-model auto-compaction budgets with native exact ids", () => {
+    writeConfig({
+      port: 10100,
+      providers: {
+        custom: {
+          adapter: "openai-chat",
+          baseUrl: "https://example.test/v1",
+          modelAutoCompactTokenLimits: { model: 1.5 },
+        },
+      },
+      defaultProvider: "custom",
+    });
+    expect(readConfigDiagnostics().source).toBe("fallback");
+    expect(readConfigDiagnostics().error).toContain("providers.custom.modelAutoCompactTokenLimits");
+
+    rmSync(testDir, { recursive: true, force: true });
+    mkdirSync(testDir, { recursive: true });
+    writeConfig({
+      port: 10100,
+      providers: {
+        openai: {
+          adapter: "openai-responses",
+          baseUrl: "https://chatgpt.com/backend-api/codex",
+          authMode: "forward",
+          codexAccountMode: "direct",
+          modelAutoCompactTokenLimits: { "team/gpt-5.6-sol": 64_000 },
+        },
+      },
+      defaultProvider: "openai",
+    });
+    expect(readConfigDiagnostics().source).toBe("fallback");
+    expect(readConfigDiagnostics().error).toContain("exact supported native model id");
+  });
+
   test("disk config preserves valid OpenRouter routing and rejects invalid destinations", () => {
     writeConfig({
       port: 10100,
diff --git a/tests/generic-oauth-failover.test.ts b/tests/generic-oauth-failover.test.ts
index 2d17f8ceb4..18b4040412 100644
--- a/tests/generic-oauth-failover.test.ts
+++ b/tests/generic-oauth-failover.test.ts
@@ -220,12 +220,15 @@ describe("sidecar on429 wiring", () => {
 
   test("every rotation site applies the credential through the one shared helper", () => {
     // The pairing rules (Copilot's account-scoped origin, Antigravity's account-matched project,
-    // Kiro's routing metadata) live in exactly one place. A fourth rotation site that swaps the
-    // bearer by hand would reintroduce the mixed-identity bug this helper exists to prevent.
-    const snapshotUses = coreSource.match(/failoverAccountSnapshot\(/g) ?? [];
+    // Kiro's routing metadata) live in exactly one place. Count both the generic snapshot resolver
+    // and the opt-in Antigravity pool resolver: adding a specialized rotation path must not create
+    // a second place that writes the bearer or its account-scoped metadata by hand.
+    const genericSnapshotUses = coreSource.match(/failoverAccountSnapshot\(/g) ?? [];
+    const googleSnapshotUses = coreSource.match(/getGoogleAntigravityPoolAccessSnapshot\(/g) ?? [];
     const helperUses = coreSource.match(/applyFailoverSnapshot\(snapshot\)/g) ?? [];
-    expect(snapshotUses.length).toBe(3);
-    expect(helperUses.length).toBe(snapshotUses.length);
+    expect(genericSnapshotUses.length).toBe(3);
+    expect(googleSnapshotUses.length).toBe(3);
+    expect(helperUses.length).toBe(genericSnapshotUses.length + googleSnapshotUses.length);
     // The bearer is written in exactly one place — inside the helper. Any other occurrence is a
     // rotation site that skipped the pairing rules.
     const bearerWrites = coreSource.match(/apiKey: snapshot\.accessToken/g) ?? [];
diff --git a/tests/state-store-sweeper.test.ts b/tests/state-store-sweeper.test.ts
index ea2d473b27..251d77e2d9 100644
--- a/tests/state-store-sweeper.test.ts
+++ b/tests/state-store-sweeper.test.ts
@@ -100,6 +100,7 @@ describe("state-store sweeper", () => {
       "provider-request-pacing",
       "combo-target-cooldowns",
       "anthropic-routing-health",
+      "google-antigravity-routing-health",
       "xai-refresh-verdicts",
       "responses-continuation",
       "antigravity-replay",

From f7b3320c039b91b8ef2d32a792793e3d625b60ec Mon Sep 17 00:00:00 2001
From: roy6732856 <roy6732856@gmail.com>
Date: Wed, 26 Aug 2026 16:36:03 +0800
Subject: [PATCH 3/5] docs(google): clarify pool recovery scope

---
 .../fr/reference/configuration/providers.md   |  6 ++++
 .../ja/reference/configuration/providers.md   |  5 +++
 .../ko/reference/configuration/providers.md   |  5 +++
 .../docs/reference/cli/providers-accounts.md  |  4 +--
 .../docs/reference/configuration/providers.md |  6 ++++
 .../ru/reference/configuration/providers.md   |  6 ++++
 .../tr/reference/configuration/providers.md   |  5 +++
 .../reference/configuration/providers.md      |  4 +++
 .../reference/configuration/providers.md      |  4 +++
 .../google-antigravity-docs-contract.test.ts  | 36 +++++++++++++++++++
 10 files changed, 79 insertions(+), 2 deletions(-)
 create mode 100644 tests/google-antigravity-docs-contract.test.ts

diff --git a/docs-site/src/content/docs/fr/reference/configuration/providers.md b/docs-site/src/content/docs/fr/reference/configuration/providers.md
index e55b7c3195..f9cceeaf85 100644
--- a/docs-site/src/content/docs/fr/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/fr/reference/configuration/providers.md
@@ -243,6 +243,12 @@ exploitable, la temporisation par défaut est de 60 secondes ; une valeur amont
 les comptes admissibles sont en temporisation, le client reçoit un 429 avec le premier `Retry-After`
 connu.
 
+La rotation spécialisée sur échec 402/429 est limitée au chemin d'envoi principal Responses ordinaire
+et aux continuations de terminal. Lorsqu'elles passent par Google, les boucles `image/video bridge` et
+`web-search sidecar` peuvent utiliser le compte initialement sélectionné pour la requête, mais elles ne
+le placent pas en temporisation et ne le font pas tourner via ce pool spécialisé. Le point de terminaison
+d'image Antigravity autonome reste lui aussi hors de ce chemin de rotation.
+
 :::caution[Résilience opérationnelle, pas contournement de quota]
 Utilisez ce pool pour récupérer après des défaillances transitoires, pas pour contourner les quotas ou
 les contrôles du fournisseur. L'automatisation multicomptes peut enfreindre les conditions du fournisseur ;
diff --git a/docs-site/src/content/docs/ja/reference/configuration/providers.md b/docs-site/src/content/docs/ja/reference/configuration/providers.md
index 3da7f62157..6d0bf659ad 100644
--- a/docs-site/src/content/docs/ja/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ja/reference/configuration/providers.md
@@ -201,6 +201,11 @@ Google AI Studio、Vertex AI、別のプロバイダー項目、API キールー
 1 リクエストにつきフェイルオーバーは最大 3 回、上流への送信は合計最大 4 回です。すべての適格
 アカウントがクールダウン中なら、クライアントには最も早い既知の `Retry-After` を付けて 429 を返します。
 
+専用の 402/429 障害ローテーションは、通常の Responses メイン送信パスとターミナル継続パスに限定されます。
+Google にルーティングされる場合、`image/video bridge` と `web-search sidecar` のループはリクエストで
+最初に選択されたアカウントを使用できますが、この専用プールを通じてそのアカウントをクールダウンまたは
+ローテーションしません。スタンドアロンの Antigravity 画像エンドポイントも、このローテーションパスの対象外です。
+
 :::caution[運用上の回復力であり、クォータ回避ではありません]
 このプールは一時的なアカウント障害からの回復に使い、クォータやプロバイダーの制限を回避するために
 使わないでください。複数アカウントの自動運用は利用規約に違反する可能性があります。そのリスクを
diff --git a/docs-site/src/content/docs/ko/reference/configuration/providers.md b/docs-site/src/content/docs/ko/reference/configuration/providers.md
index 61eaf5dbae..7509629f81 100644
--- a/docs-site/src/content/docs/ko/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ko/reference/configuration/providers.md
@@ -207,6 +207,11 @@ Vertex AI, 다른 공급자 항목 또는 API 키 경로에 사용되지 않습
 총 4회까지 허용됩니다. 모든 적격 계정이 쿨다운 중이면 가장 이른 알려진 `Retry-After`와 함께 429를
 클라이언트에 반환합니다.
 
+특수 402/429 장애 순환은 일반 Responses 주 전송 경로와 터미널 연속 경로로 제한됩니다. Google로
+라우팅될 때 `image/video bridge`와 `web-search sidecar` 루프는 요청에서 처음 선택한 계정을 사용할 수
+있지만, 이 특수 풀을 통해 해당 계정을 쿨다운하거나 순환하지 않습니다. 독립형 Antigravity 이미지
+엔드포인트도 이 순환 경로의 범위 밖입니다.
+
 :::caution[운영 복원력이며 할당량 우회가 아닙니다]
 이 풀은 일시적인 계정 장애를 복구하는 데 사용하고 할당량이나 공급자 제한을 회피하는 데 사용하지
 마십시오. 다중 계정 자동화는 공급자 약관을 위반할 수 있으므로 그 위험을 감수하지 않는다면 비활성화
diff --git a/docs-site/src/content/docs/reference/cli/providers-accounts.md b/docs-site/src/content/docs/reference/cli/providers-accounts.md
index 80f1172260..0345266d8f 100644
--- a/docs-site/src/content/docs/reference/cli/providers-accounts.md
+++ b/docs-site/src/content/docs/reference/cli/providers-accounts.md
@@ -161,8 +161,8 @@ returns:
 ```
 
 `--quota` adds a `QUOTA` column with each account's own usage, for providers that support a
-per-account probe (Anthropic today). It is opt-in because the proxy probes the upstream once per
-stored credential; the default listing stays a local read. `--refresh` bypasses the cached
+per-account probe (Anthropic and Google Antigravity today). It is opt-in because the proxy probes
+the upstream once per stored credential; the default listing stays a local read. `--refresh` bypasses the cached
 result. An account with no per-account quota shows `-`, and one whose probe failed shows
 `unavailable` — blank would read as "no usage" rather than "not measured". `--json` carries the
 full breakdown per account, not just the two summarized windows:
diff --git a/docs-site/src/content/docs/reference/configuration/providers.md b/docs-site/src/content/docs/reference/configuration/providers.md
index 9beb09b81c..5fed411f5a 100644
--- a/docs-site/src/content/docs/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/reference/configuration/providers.md
@@ -326,6 +326,12 @@ rebuilds the request for another eligible snapshot. The default cooldown is 60 s
 three failovers (four total upstream dispatches). If every eligible account is cooling, the client
 receives 429 with the earliest known `Retry-After`.
 
+Specialized 402/429 failure rotation is limited to the ordinary Responses main dispatch and terminal
+continuation paths. When routed through Google, the `image/video bridge` and `web-search sidecar`
+loops can use the account initially selected for the request, but they do not cool or rotate it
+through this specialized pool. The standalone Antigravity image endpoint is also outside this
+rotation path.
+
 :::caution[Operational resilience, not quota bypass]
 Use this pool to recover from transient account failures, not to evade quotas or provider enforcement.
 Multi-account automation can violate provider terms; keep the feature disabled unless you accept that
diff --git a/docs-site/src/content/docs/ru/reference/configuration/providers.md b/docs-site/src/content/docs/ru/reference/configuration/providers.md
index 27d552be06..eb7376a17e 100644
--- a/docs-site/src/content/docs/ru/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ru/reference/configuration/providers.md
@@ -246,6 +246,12 @@ cooldown, очищает его привязку и пересобирает з
 15 минутами. Один запрос допускает не более трёх failover, то есть всего четырёх upstream dispatch.
 Если все подходящие аккаунты в cooldown, клиент получает 429 с самым ранним известным `Retry-After`.
 
+Специализированная ротация при 402/429 ограничена обычным основным dispatch-путём Responses и terminal
+continuation. При маршрутизации через Google циклы `image/video bridge` и `web-search sidecar` могут
+использовать аккаунт, изначально выбранный для запроса, но не переводят его в cooldown и не ротируют
+через этот специализированный пул. Отдельный endpoint изображений Antigravity также находится вне
+этого пути ротации.
+
 :::caution[Операционная устойчивость, а не обход квот]
 Используйте пул для восстановления после временных сбоев аккаунта, а не для обхода квот или enforcement
 провайдера. Автоматизация нескольких аккаунтов может нарушать условия провайдера; оставьте функцию
diff --git a/docs-site/src/content/docs/tr/reference/configuration/providers.md b/docs-site/src/content/docs/tr/reference/configuration/providers.md
index 6c2110b7af..32d8cf22e1 100644
--- a/docs-site/src/content/docs/tr/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/tr/reference/configuration/providers.md
@@ -265,6 +265,11 @@ bağlılığını temizler ve isteği başka bir uygun anlık görüntü için y
 Bir istek en fazla üç yük devretmeye, yani toplam dört upstream gönderime izin verir. Uygun hesapların
 tamamı soğuyorsa istemci en erken bilinen `Retry-After` ile 429 alır.
 
+Özelleşmiş 402/429 hata rotasyonu, normal Responses ana gönderim ve terminal devam yollarıyla sınırlıdır.
+Google üzerinden yönlendirildiklerinde `image/video bridge` ve `web-search sidecar` döngüleri isteğin
+başlangıçta seçilen hesabını kullanabilir, ancak bu özelleşmiş havuz üzerinden hesabı soğumaya almaz veya
+döndürmez. Bağımsız Antigravity görüntü uç noktası da bu rotasyon yolunun dışındadır.
+
 :::caution[Operasyonel dayanıklılık; kota aşma yöntemi değildir]
 Bu havuzu geçici hesap hatalarından kurtulmak için kullanın; kota veya sağlayıcı yaptırımını aşmak için
 kullanmayın. Çoklu hesap otomasyonu sağlayıcı koşullarını ihlal edebilir; bu riski kabul etmiyorsanız
diff --git a/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md b/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
index a54afdac23..9211e8b6d2 100644
--- a/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
@@ -194,6 +194,10 @@ Assist `projectId`。最终 429 或透传到此处的 402 会让失败账户进
 15 分钟。单个请求最多允许 3 次故障转移，即总计 4 次上游派发。如果所有合格账户都在冷却，
 客户端会收到 429，并带上已知的最早 `Retry-After`。
 
+专用的 402/429 故障轮换仅适用于普通 Responses 主派发路径和终端续接路径。通过 Google 路由时，
+`image/video bridge` 和 `web-search sidecar` 循环可以使用请求最初选定的账户，但不会通过此专用池
+让该账户进入冷却或进行轮换。独立的 Antigravity 图像端点同样不在此轮换路径内。
+
 :::caution[用于运行韧性，而不是绕过配额]
 请只用该池从暂时性账户故障中恢复，不要用它规避配额或提供方管控。多账户自动化可能违反提供方
 条款；如果不接受这一风险，请保持关闭。
diff --git a/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md b/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
index a2d5354352..63da8d1f03 100644
--- a/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
@@ -163,6 +163,10 @@ Assist `projectId`。最終 429 或傳到此處的 402 會讓失敗帳號進入
 15 分鐘。單一請求最多允許 3 次容錯移轉，也就是總計 4 次上游派送。若所有合格帳號都在冷卻，
 客戶端會收到 429，並附上已知最早的 `Retry-After`。
 
+專用的 402/429 故障輪換只適用於一般 Responses 主派送路徑與終端續接路徑。透過 Google 路由時，
+`image/video bridge` 與 `web-search sidecar` 迴圈可以使用請求最初選定的帳號，但不會透過此專用池
+讓該帳號進入冷卻或輪換。獨立的 Antigravity 圖像端點也不在此輪換路徑內。
+
 :::caution[用於營運韌性，而非繞過配額]
 請只用此池從暫時性帳號故障中復原，不要用來規避配額或供應商管控。多帳號自動化可能違反供應商
 條款；若不接受這項風險，請保持停用。
diff --git a/tests/google-antigravity-docs-contract.test.ts b/tests/google-antigravity-docs-contract.test.ts
new file mode 100644
index 0000000000..38d8ef6e8a
--- /dev/null
+++ b/tests/google-antigravity-docs-contract.test.ts
@@ -0,0 +1,36 @@
+import { describe, expect, test } from "bun:test";
+
+const LOCALES = ["", "fr", "ja", "ko", "ru", "tr", "zh-cn", "zh-tw"] as const;
+
+function providersReferencePath(locale: string): string {
+  const prefix = locale ? `${locale}/` : "";
+  return `../docs-site/src/content/docs/${prefix}reference/configuration/providers.md`;
+}
+
+function headingSection(text: string, heading: string): string {
+  const start = text.indexOf(heading);
+  expect(start).toBeGreaterThanOrEqual(0);
+  const body = text.slice(start + heading.length);
+  const nextHeading = body.indexOf("\n### ");
+  return nextHeading === -1 ? body : body.slice(0, nextHeading);
+}
+
+describe("Google Antigravity documentation contract", () => {
+  test("every provider reference scopes specialized failure rotation away from sidecar loops", async () => {
+    for (const locale of LOCALES) {
+      const text = await Bun.file(new URL(providersReferencePath(locale), import.meta.url)).text();
+      const section = headingSection(text, "### `googleAntigravityAccountPool`");
+      expect(section, providersReferencePath(locale)).toContain("image/video bridge");
+      expect(section, providersReferencePath(locale)).toContain("web-search sidecar");
+      expect(section, providersReferencePath(locale)).toContain("402/429");
+    }
+  });
+
+  test("account quota docs name both providers with per-account probes", async () => {
+    const text = await Bun.file(new URL(
+      "../docs-site/src/content/docs/reference/cli/providers-accounts.md",
+      import.meta.url,
+    )).text();
+    expect(text).toContain("per-account probe (Anthropic and Google Antigravity today)");
+  });
+});

From 92dc5554780937fb31dea57ba365a7347c497dc7 Mon Sep 17 00:00:00 2001
From: roy6732856 <roy6732856@gmail.com>
Date: Wed, 26 Aug 2026 20:28:42 +0800
Subject: [PATCH 4/5] fix(google): share dispatch cap with continuation

---
 src/server/responses/core.ts                  | 51 ++++++++++++++-
 ...e-antigravity-account-pool-request.test.ts | 65 ++++++++++++++++---
 2 files changed, 105 insertions(+), 11 deletions(-)

diff --git a/src/server/responses/core.ts b/src/server/responses/core.ts
index ddd19e3562..64466aa74f 100644
--- a/src/server/responses/core.ts
+++ b/src/server/responses/core.ts
@@ -854,6 +854,8 @@ function assertNeverGoogleAntigravityRotation(value: never): never {
 
 const GOOGLE_ANTIGRAVITY_FAILOVER_LIMIT_MESSAGE =
   "Google Antigravity quota exhausted after bounded account failover";
+const GOOGLE_ANTIGRAVITY_POOL_MAX_DISPATCHES_PER_REQUEST =
+  GOOGLE_ANTIGRAVITY_POOL_MAX_FAILOVERS_PER_REQUEST + 1;
 
 function googleAntigravityFailoverLimitResponse(status: 429 | 402, retryAfter: string | null): Response {
   return formatErrorResponse(
@@ -2891,6 +2893,7 @@ async function handleResponsesInner(
     : null;
   let googleAntigravityPoolAccountId: string | null = null;
   let googleAntigravityPoolFailovers = 0;
+  let googleAntigravityPoolDispatches = 0;
   const googleAntigravityPoolActive = route.providerName === "google-antigravity"
     && route.provider.authMode === "oauth"
     && route.provider.googleMode === "cloud-code-assist"
@@ -4968,6 +4971,7 @@ async function handleResponsesInner(
   const invalidateSameTargetRequest = (): void => { transportToken += 1; };
   let upstreamResponse: Response;
   try {
+    if (googleAntigravityPoolActive) googleAntigravityPoolDispatches += 1;
     if (activeAdapter.fetchResponse) {
       noteAttemptSend(logCtx.activeAttempt, inputTokenEstimate);
       await waitForProviderRequestSlot(route.providerName, route.provider, route.modelId, upstream.signal);
@@ -5039,6 +5043,12 @@ async function handleResponsesInner(
     const rebuildAndRefetch = async (
       recovery: AttemptRecoveryKind,
     ): Promise<Response | { failed: Response }> => {
+      if (
+        googleAntigravityPoolActive
+        && googleAntigravityPoolDispatches >= GOOGLE_ANTIGRAVITY_POOL_MAX_DISPATCHES_PER_REQUEST
+      ) {
+        return { failed: googleAntigravityFailoverLimitResponse(429, null) };
+      }
       let retryRequest: AdapterRequest;
       if (sameTargetRequest !== undefined && sameTargetParsed === parsed && sameTargetToken === transportToken) {
         // Same target (key/adapter/parsed/tier unchanged): replay the exact cached request.
@@ -5074,6 +5084,7 @@ async function handleResponsesInner(
       logCtx.providerAdapter = activeAdapter.name;
       sealRequestAttemptIdentity(logCtx.activeAttempt, logCtx.provider, activeAdapter.name, logCtx.accountLogLabel);
       noteAttemptSend(logCtx.activeAttempt, retryEstimate, recovery);
+      if (googleAntigravityPoolActive) googleAntigravityPoolDispatches += 1;
       try {
         try {
           if (activeAdapter.fetchResponse) {
@@ -5289,7 +5300,10 @@ async function handleResponsesInner(
           case "no-eligible-account":
             break googleAntigravityFailover;
           case "next-account":
-            if (googleAntigravityPoolFailovers >= GOOGLE_ANTIGRAVITY_POOL_MAX_FAILOVERS_PER_REQUEST) {
+            if (
+              googleAntigravityPoolFailovers >= GOOGLE_ANTIGRAVITY_POOL_MAX_FAILOVERS_PER_REQUEST
+              || googleAntigravityPoolDispatches >= GOOGLE_ANTIGRAVITY_POOL_MAX_DISPATCHES_PER_REQUEST
+            ) {
               try { void upstreamResponse.body?.cancel().catch(() => {}); } catch { /* already consumed/closed */ }
               return googleAntigravityFailoverLimitResponse(
                 upstreamResponse.status === 402 ? 402 : 429,
@@ -5555,6 +5569,7 @@ async function handleResponsesInner(
       // Optional recovery label for same-target / failover continuation sends.
       const replayKind: AttemptRecoveryKind | undefined = recoveryKind;
       try {
+        if (googleAntigravityPoolActive) googleAntigravityPoolDispatches += 1;
         if (activeAdapter.fetchResponse) {
           noteAttemptSend(logCtx.activeAttempt, continuationEstimate, replayKind);
           await waitForProviderRequestSlot(route.providerName, route.provider, nextParsed.modelId, upstream.signal);
@@ -5597,6 +5612,19 @@ async function handleResponsesInner(
       }
     };
     while (true) {
+      if (
+        googleAntigravityPoolActive
+        && googleAntigravityPoolDispatches >= GOOGLE_ANTIGRAVITY_POOL_MAX_DISPATCHES_PER_REQUEST
+      ) {
+        googleAntigravityContinuationFailoverLimit.value = { status: 429, retryAfter: null };
+        yield {
+          type: "error",
+          status: 429,
+          errorType: "rate_limit_error",
+          message: GOOGLE_ANTIGRAVITY_FAILOVER_LIMIT_MESSAGE,
+        };
+        return;
+      }
       try {
         const recoveryKind = nextContinuationRecoveryKind;
         nextContinuationRecoveryKind = undefined;
@@ -5645,6 +5673,22 @@ async function handleResponsesInner(
           yield { type: "error", message: "client closed request during terminal continuation", status: 499 };
           return;
         }
+        if (
+          googleAntigravityPoolActive
+          && googleAntigravityPoolDispatches >= GOOGLE_ANTIGRAVITY_POOL_MAX_DISPATCHES_PER_REQUEST
+        ) {
+          googleAntigravityContinuationFailoverLimit.value = {
+            status: 429,
+            retryAfter: response.headers.get("retry-after"),
+          };
+          yield {
+            type: "error",
+            status: 429,
+            errorType: "rate_limit_error",
+            message: GOOGLE_ANTIGRAVITY_FAILOVER_LIMIT_MESSAGE,
+          };
+          return;
+        }
         try {
           response = await fetchContinuation("rate-limit-429");
         } catch (error) {
@@ -5753,7 +5797,10 @@ async function handleResponsesInner(
           case "no-eligible-account":
             break;
           case "next-account": {
-            if (googleAntigravityPoolFailovers >= GOOGLE_ANTIGRAVITY_POOL_MAX_FAILOVERS_PER_REQUEST) {
+            if (
+              googleAntigravityPoolFailovers >= GOOGLE_ANTIGRAVITY_POOL_MAX_FAILOVERS_PER_REQUEST
+              || googleAntigravityPoolDispatches >= GOOGLE_ANTIGRAVITY_POOL_MAX_DISPATCHES_PER_REQUEST
+            ) {
               try { void response.body?.cancel().catch(() => {}); } catch { /* already consumed/closed */ }
               const status = response.status as 429 | 402;
               googleAntigravityContinuationFailoverLimit.value = {
diff --git a/tests/google-antigravity-account-pool-request.test.ts b/tests/google-antigravity-account-pool-request.test.ts
index a41a622a45..907102302f 100644
--- a/tests/google-antigravity-account-pool-request.test.ts
+++ b/tests/google-antigravity-account-pool-request.test.ts
@@ -259,8 +259,8 @@ describe("google antigravity Responses account failover", () => {
     }
   });
 
-  test.each([1, 2, 3, 4])(
-    "terminal continuation exhaustion across %d eligible accounts surfaces all-cooled",
+  test.each([1, 2, 3])(
+    "terminal continuation exhaustion across %d eligible accounts surfaces all-cooled within the dispatch cap",
     async eligibleCount => {
     const { accountIds } = await seedAccounts(eligibleCount, 4);
     let initialSends = 0;
@@ -321,8 +321,10 @@ describe("google antigravity Responses account failover", () => {
     },
   );
 
-  test("terminal continuation cap preserves quota status without dispatching an eligible fifth account", async () => {
-    const { accountIds } = await seedAccounts(5, 5);
+  test.each([4, 5])(
+    "terminal continuation cap across %d eligible accounts stops at four total dispatches",
+    async eligibleCount => {
+    const { accountIds } = await seedAccounts(eligibleCount, 5);
     let initialSends = 0;
     const quotaDispatches: Dispatch[] = [];
     const upstream = Bun.serve({
@@ -351,20 +353,65 @@ describe("google antigravity Responses account failover", () => {
     try {
       const response = await fetch(new URL("/v1/responses", proxy.url), {
         method: "POST",
-        headers: { "content-type": "application/json", "thread-id": "five-account-continuation-cap" },
+        headers: { "content-type": "application/json", "thread-id": `continuation-cap-${eligibleCount}` },
         body: JSON.stringify({ model: "gemini-3.7-flash", input: "hello", stream: false }),
       });
       const body = await response.text();
 
-      expect([response.status, response.headers.get("retry-after")]).toEqual([402, "450"]);
+      expect([response.status, response.headers.get("retry-after")]).toEqual([429, "400"]);
       expect(body).toContain("Google Antigravity quota exhausted after bounded account failover");
       expect(initialSends).toBe(1);
-      expect(quotaDispatches).toEqual(["a", "b", "c", "d"].map(suffix => ({
+      expect(quotaDispatches).toEqual(["a", "b", "c"].map(suffix => ({
+        authorization: `Bearer request-token-${suffix}`,
+        project: `request-project-${suffix}`,
+      })));
+      expect(accountIds.slice(0, 3).every(id => getGoogleAntigravityAccountHealthSnapshot(id) !== null)).toBe(true);
+      expect(accountIds.slice(3).every(id => getGoogleAntigravityAccountHealthSnapshot(id) === null)).toBe(true);
+    } finally {
+      await proxy.stop(true);
+      upstream.stop(true);
+    }
+    },
+  );
+
+  test("main failovers consume the shared dispatch budget before terminal continuation", async () => {
+    await seedAccounts(5, 5);
+    const dispatches: Dispatch[] = [];
+    const upstream = Bun.serve({
+      port: 0,
+      hostname: "127.0.0.1",
+      async fetch(request) {
+        const body = await request.json() as { project?: unknown };
+        const authorization = request.headers.get("authorization");
+        dispatches.push({ authorization, project: body.project });
+        if (authorization !== "Bearer request-token-d") {
+          return exhaustedCcaResponse(authorization);
+        }
+        return Response.json({
+          response: {
+            candidates: [{ content: { parts: [] }, finishReason: "STOP" }],
+            usageMetadata: { promptTokenCount: 1, candidatesTokenCount: 0 },
+          },
+        });
+      },
+    });
+    saveConfig({ ...config(upstream.url.origin), emptyCompletionRetry: true });
+    const proxy = startServer(0);
+    try {
+      const response = await fetch(new URL("/v1/responses", proxy.url), {
+        method: "POST",
+        headers: { "content-type": "application/json", "thread-id": "main-exhausts-shared-cap" },
+        body: JSON.stringify({ model: "gemini-3.7-flash", input: "hello", stream: false }),
+      });
+      const body = await response.text();
+
+      expect(response.status).toBe(429);
+      expect(response.headers.get("retry-after")).toBeNull();
+      expect(body).toContain("Google Antigravity quota exhausted after bounded account failover");
+      expect(dispatches).toEqual(["a", "b", "c", "d"].map(suffix => ({
         authorization: `Bearer request-token-${suffix}`,
         project: `request-project-${suffix}`,
       })));
-      expect(accountIds.slice(0, 4).every(id => getGoogleAntigravityAccountHealthSnapshot(id) !== null)).toBe(true);
-      expect(getGoogleAntigravityAccountHealthSnapshot(accountIds[4]!)).toBeNull();
     } finally {
       await proxy.stop(true);
       upstream.stop(true);

From 4bab2fbbc8830bf18c28e04132d434314c09566d Mon Sep 17 00:00:00 2001
From: roy6732856 <roy6732856@gmail.com>
Date: Wed, 26 Aug 2026 21:05:38 +0800
Subject: [PATCH 5/5] fix(google): make pool off strictly single-account

---
 .../fr/reference/configuration/providers.md   | 15 +++----
 .../ja/reference/configuration/providers.md   | 11 +++--
 .../ko/reference/configuration/providers.md   | 14 +++----
 .../docs/reference/configuration/providers.md | 11 +++--
 .../ru/reference/configuration/providers.md   | 16 +++----
 .../tr/reference/configuration/providers.md   | 12 ++++--
 .../reference/configuration/providers.md      | 10 +++--
 .../reference/configuration/providers.md      | 10 +++--
 src/server/management/oauth-account-routes.ts | 23 ++++++----
 tests/account-pool-management-api.test.ts     | 28 ++++++++++++-
 .../google-antigravity-docs-contract.test.ts  | 42 +++++++++++++++++++
 11 files changed, 142 insertions(+), 50 deletions(-)

diff --git a/docs-site/src/content/docs/fr/reference/configuration/providers.md b/docs-site/src/content/docs/fr/reference/configuration/providers.md
index f9cceeaf85..94f69a7ceb 100644
--- a/docs-site/src/content/docs/fr/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/fr/reference/configuration/providers.md
@@ -213,13 +213,14 @@ Cette option regroupe des comptes OAuth Google Antigravity uniquement pour le fo
 Assist `google-antigravity`. Elle est désactivée par défaut. Le pool ne fournit jamais d'identifiants à
 Google AI Studio, Vertex AI, une autre entrée de fournisseur ou une route utilisant une clé API.
 
-Désactiver ce pool spécialisé désactive la sélection selon les quotas, l'affinité de session et son
-budget de nouvelles tentatives après un 402 ou un 429. Cela ne désactive pas le mécanisme distinct de
-basculement OAuth générique sur 429, activé en fonction de la présence de comptes. Lorsque
-`googleAntigravityAccountPool.enabled` vaut `true`, le pool spécialisé prend en charge le routage
-Google ; lorsqu'il est désactivé, la stratégie générique peut encore assurer un basculement d'urgence
-après un 429. Pour limiter strictement Google à un seul compte, définissez également
-`providers.google-antigravity.oauthAccountFailover.enabled` sur `false`.
+Définir directement `googleAntigravityAccountPool.enabled` sur `false` dans la configuration désactive
+la sélection selon les quotas, l'affinité de session et le budget spécialisé 402/429 sans modifier le
+basculement OAuth générique distinct. En revanche, `ocx account auto-switch google-antigravity off` et un
+`PUT/PATCH /api/oauth/accounts/pool` de l'API de gestion avec `enabled: false` explicite imposent le
+mode strict à un seul compte : ils enregistrent aussi
+`providers.google-antigravity.oauthAccountFailover.enabled` à `false`. Une mise à jour partielle qui
+omet `enabled` conserve l'intention existante du basculement générique. Pour garder ce basculement avec
+le pool spécialisé désactivé, modifiez directement la configuration et réglez cet override sur `true`.
 
 | Clé | Type | Par défaut | Description |
 | --- | --- | --- | --- |
diff --git a/docs-site/src/content/docs/ja/reference/configuration/providers.md b/docs-site/src/content/docs/ja/reference/configuration/providers.md
index 6d0bf659ad..2b8273d452 100644
--- a/docs-site/src/content/docs/ja/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ja/reference/configuration/providers.md
@@ -175,10 +175,13 @@ Anthropic アカウント ポリシーのリスクを理解していない限り
 Antigravity OAuth アカウントをプールします。デフォルトでは無効です。このプールの認証情報が
 Google AI Studio、Vertex AI、別のプロバイダー項目、API キールートに使われることはありません。
 
-この専用プールを無効にすると、クォータを考慮した選択、セッション親和性、および専用の 402/429
-再試行予算が無効になります。ただし、アカウントの存在によって有効になる別個の汎用 OAuth 429
-フォールバックは無効になりません。Google を厳密に単一アカウントで運用するには、
-`providers.google-antigravity.oauthAccountFailover.enabled` も `false` に設定してください。
+設定で `googleAntigravityAccountPool.enabled` を直接 `false` にすると、クォータを考慮した選択、
+セッション親和性、専用の 402/429 再試行予算だけが無効になり、別個の汎用 OAuth fallback は変更されません。
+一方、`ocx account auto-switch google-antigravity off` と、`enabled: false` を明示した Management API の
+`PUT/PATCH /api/oauth/accounts/pool` は strict single-account 操作であり、
+`providers.google-antigravity.oauthAccountFailover.enabled` も `false` として保存します。
+`enabled` を省略した部分更新は既存の汎用 failover の意図を保持します。専用プールを無効にしたまま
+汎用 fallback を維持するには、設定を直接編集してこの provider override を `true` にしてください。
 
 | キー | 型 | デフォルト | 説明 |
 | --- | --- | --- | --- |
diff --git a/docs-site/src/content/docs/ko/reference/configuration/providers.md b/docs-site/src/content/docs/ko/reference/configuration/providers.md
index 7509629f81..cccf9493b2 100644
--- a/docs-site/src/content/docs/ko/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ko/reference/configuration/providers.md
@@ -179,13 +179,13 @@ Anthropic 계정 정책 위험을 이해하지 못한다면 이 기능은 꺼두
 계정을 풀로 묶습니다. 기본값은 비활성화입니다. 이 풀의 자격 증명은 Google AI Studio,
 Vertex AI, 다른 공급자 항목 또는 API 키 경로에 사용되지 않습니다.
 
-이 특수 풀을 비활성화하면 할당량 인식 선택, 세션 affinity, 해당 풀의 402/429 재시도
-예산이 꺼집니다. 그러나 별도로 계정 존재 여부에 따라 활성화되는 일반 OAuth 429 fallback은
-비활성화되지 않습니다. Google을 엄격하게 단일 계정으로 유지하려면
-`providers.google-antigravity.oauthAccountFailover.enabled`도 `false`로 설정하십시오.
-
-`googleAntigravityAccountPool.enabled`가 `true`이면 이 특수 풀이 Google 라우팅을 소유합니다.
-꺼져 있으면 일반 `oauthAccountFailover` 정책이 여전히 긴급 429 failover를 제공할 수 있습니다.
+설정에서 `googleAntigravityAccountPool.enabled`를 직접 `false`로 지정하면 할당량 인식 선택,
+세션 affinity, 특수 402/429 재시도 예산만 꺼지고 별도의 일반 OAuth fallback은 변경되지 않습니다.
+반면 `ocx account auto-switch google-antigravity off`와 `enabled: false`를 명시한 Management API의
+`PUT/PATCH /api/oauth/accounts/pool`은 strict single-account 제어입니다. 이 경로는
+`providers.google-antigravity.oauthAccountFailover.enabled`도 `false`로 저장합니다.
+`enabled`를 생략한 부분 업데이트는 기존 일반 failover 의도를 유지합니다. 특수 풀을 끈 채 일반
+fallback을 유지하려면 설정을 직접 편집하여 이 공급자 override를 `true`로 지정하십시오.
 
 | 키 | 타입 | 기본값 | 설명 |
 | --- | --- | --- | --- |
diff --git a/docs-site/src/content/docs/reference/configuration/providers.md b/docs-site/src/content/docs/reference/configuration/providers.md
index 5fed411f5a..ac874f1939 100644
--- a/docs-site/src/content/docs/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/reference/configuration/providers.md
@@ -301,10 +301,13 @@ This opt-in pools Google Antigravity OAuth accounts for the `google-antigravity`
 provider only. It is disabled by default. The pool never supplies credentials to Google AI Studio,
 Vertex AI, another provider entry, or an API-key route.
 
-Disabling this specialized pool turns off quota-aware selection, session affinity, and its 402/429
-retry budget. It does not disable the separate presence-driven generic OAuth 429 fallback. To keep
-Google strictly on one account, also set
-`providers.google-antigravity.oauthAccountFailover.enabled` to `false`.
+Setting `googleAntigravityAccountPool.enabled` to `false` directly in config disables quota-aware
+selection, session affinity, and the specialized 402/429 retry budget without changing the separate
+generic OAuth fallback. By contrast, `ocx account auto-switch google-antigravity off` and a Management API
+`PUT/PATCH /api/oauth/accounts/pool` with explicit `enabled: false` are strict single-account controls:
+they also persist `providers.google-antigravity.oauthAccountFailover.enabled` as `false`. Partial
+updates that omit `enabled` preserve the existing generic-failover intent. To keep generic fallback
+while the specialized pool is off, edit config directly and set that provider override to `true`.
 
 | Key | Type | Default | Description |
 | --- | --- | --- | --- |
diff --git a/docs-site/src/content/docs/ru/reference/configuration/providers.md b/docs-site/src/content/docs/ru/reference/configuration/providers.md
index eb7376a17e..bed98cdb4a 100644
--- a/docs-site/src/content/docs/ru/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/ru/reference/configuration/providers.md
@@ -216,14 +216,14 @@ backoff и может переключить аккаунт уже внутри
 провайдера `google-antigravity`. По умолчанию она выключена. Пул никогда не передаёт credentials в
 Google AI Studio, Vertex AI, другую запись провайдера или маршрут с API-ключом.
 
-Отключение этого специализированного пула выключает выбор с учётом квоты, привязку сессий и его
-бюджет повторов для 402/429. Оно не отключает отдельный generic OAuth-failover при 429, который
-активируется наличием аккаунтов. Чтобы Google строго использовал один аккаунт, также задайте
-`providers.google-antigravity.oauthAccountFailover.enabled` равным `false`.
-
-Когда `googleAntigravityAccountPool.enabled` равен `true`, маршрутизацией Google владеет этот
-специализированный пул. Когда он выключен, отдельная generic policy всё ещё может выполнять
-аварийный failover при 429.
+Если напрямую задать `googleAntigravityAccountPool.enabled` равным `false` в конфигурации, отключатся
+выбор с учётом квоты, привязка сессий и специализированный бюджет 402/429, но отдельный generic OAuth
+fallback не изменится. В отличие от этого, `ocx account auto-switch google-antigravity off` и Management API
+`PUT/PATCH /api/oauth/accounts/pool` с явным `enabled: false` являются strict single-account операциями:
+они также сохраняют `providers.google-antigravity.oauthAccountFailover.enabled` равным `false`.
+Частичное обновление без `enabled` сохраняет текущий intent generic failover. Чтобы оставить generic
+fallback при выключенном специализированном пуле, измените конфигурацию напрямую и задайте override
+провайдера равным `true`.
 
 | Ключ | Тип | По умолчанию | Описание |
 | --- | --- | --- | --- |
diff --git a/docs-site/src/content/docs/tr/reference/configuration/providers.md b/docs-site/src/content/docs/tr/reference/configuration/providers.md
index 32d8cf22e1..63b2640824 100644
--- a/docs-site/src/content/docs/tr/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/tr/reference/configuration/providers.md
@@ -240,10 +240,14 @@ Bu isteğe bağlı özellik Google Antigravity OAuth hesaplarını yalnızca `go
 Assist sağlayıcısı için havuzlar. Varsayılan olarak devre dışıdır. Havuz, kimlik bilgilerini hiçbir
 zaman Google AI Studio, Vertex AI, başka bir sağlayıcı kaydı veya API anahtarı rotasına vermez.
 
-Bu özelleşmiş havuzu devre dışı bırakmak kota duyarlı seçimi, oturum bağlılığını ve kendi 402/429
-yeniden deneme bütçesini kapatır. Ayrı, varlığa dayalı genel OAuth 429 yük devretmesini devre dışı
-bırakmaz. Google'ı katı biçimde tek hesapta tutmak için ayrıca
-`providers.google-antigravity.oauthAccountFailover.enabled` değerini `false` olarak ayarlayın.
+Yapılandırmada `googleAntigravityAccountPool.enabled` değerini doğrudan `false` yapmak kota duyarlı
+seçimi, oturum bağlılığını ve özelleşmiş 402/429 yeniden deneme bütçesini kapatır; ayrı genel OAuth
+fallback ayarını değiştirmez. Buna karşılık `ocx account auto-switch google-antigravity off` ile açıkça
+`enabled: false` gönderen Management API `PUT/PATCH /api/oauth/accounts/pool`, strict single-account
+denetimleridir ve `providers.google-antigravity.oauthAccountFailover.enabled` değerini de `false`
+olarak kaydeder. `enabled` alanını atlayan kısmi güncellemeler mevcut genel failover niyetini korur.
+Özelleşmiş havuz kapalıyken genel fallback'i korumak için yapılandırmayı doğrudan düzenleyip bu sağlayıcı
+override değerini `true` yapın.
 
 | Anahtar | Tip | Varsayılan | Açıklama |
 | --- | --- | --- | --- |
diff --git a/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md b/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
index 9211e8b6d2..8343518622 100644
--- a/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/zh-cn/reference/configuration/providers.md
@@ -172,9 +172,13 @@ affinity。这些策略不能规避 provider enforcement。
 此可选功能仅为 `google-antigravity` Cloud Code Assist 提供方池化 Google Antigravity OAuth
 账户，默认关闭。该池绝不会把凭据用于 Google AI Studio、Vertex AI、其他提供方条目或 API 密钥路由。
 
-禁用此专用池会关闭配额感知选择、会话亲和性及其 402/429 重试预算，但不会禁用另一项由账户
-存在情况驱动的通用 OAuth 429 故障转移。要让 Google 严格只使用一个账户，还需将
-`providers.google-antigravity.oauthAccountFailover.enabled` 设为 `false`。
+在配置中直接将 `googleAntigravityAccountPool.enabled` 设为 `false`，只会关闭配额感知选择、
+会话亲和性及专用 402/429 重试预算，不会改变独立的通用 OAuth fallback。相比之下，
+`ocx account auto-switch google-antigravity off` 以及显式提交 `enabled: false` 的 Management API
+`PUT/PATCH /api/oauth/accounts/pool` 属于 strict single-account 操作：它们还会将
+`providers.google-antigravity.oauthAccountFailover.enabled` 持久化为 `false`。省略 `enabled` 的
+部分更新会保留现有的通用 failover 意图。若要在专用池关闭时保留通用 fallback，请直接编辑
+配置并将此提供方 override 设为 `true`。
 
 | 键 | 类型 | 默认值 | 说明 |
 | --- | --- | --- | --- |
diff --git a/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md b/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
index 63da8d1f03..2918ef08aa 100644
--- a/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
+++ b/docs-site/src/content/docs/zh-tw/reference/configuration/providers.md
@@ -141,9 +141,13 @@ API-key 供應商可持有字面值金鑰或環境參考。OAuth 供應商使用
 此選用功能只為 `google-antigravity` Cloud Code Assist 供應商池化 Google Antigravity OAuth
 帳號，預設停用。此池絕不會把憑證用於 Google AI Studio、Vertex AI、其他供應商項目或 API-key 路由。
 
-停用此專用池會關閉配額感知選擇、session 親和性及其 402/429 重試額度，但不會停用另一個
-由帳號存在情況驅動的通用 OAuth 429 容錯移轉。若要讓 Google 嚴格只使用單一帳號，還必須將
-`providers.google-antigravity.oauthAccountFailover.enabled` 設為 `false`。
+在設定中直接將 `googleAntigravityAccountPool.enabled` 設為 `false`，只會關閉配額感知選擇、
+session 親和性及專用 402/429 重試額度，不會改變獨立的通用 OAuth fallback。相較之下，
+`ocx account auto-switch google-antigravity off` 與明確送出 `enabled: false` 的 Management API
+`PUT/PATCH /api/oauth/accounts/pool` 都是 strict single-account 操作：它們也會將
+`providers.google-antigravity.oauthAccountFailover.enabled` 持久化為 `false`。省略 `enabled` 的
+部分更新會保留既有的通用 failover 意圖。若要在專用池關閉時保留通用 fallback，請直接編輯
+設定並將此供應商 override 設為 `true`。
 
 | Key | 型別 | 預設值 | 說明 |
 | --- | --- | --- | --- |
diff --git a/src/server/management/oauth-account-routes.ts b/src/server/management/oauth-account-routes.ts
index 2b8cdabdda..16e2ba9bfa 100644
--- a/src/server/management/oauth-account-routes.ts
+++ b/src/server/management/oauth-account-routes.ts
@@ -27,7 +27,7 @@ import { OAuthMutationBusyError, removeCredential } from "../../oauth/store";
 import { providerDestinationResolvedError } from "../../lib/destination-policy";
 import { reconcileLiveStateStores } from "../../lib/state-store-registrations";
 import { enrichProviderFromCatalog, listKeyLoginProviders } from "../../oauth/key-providers";
-import { deriveProviderPresets } from "../../providers/derive";
+import { deriveOAuthProviderConfig, deriveProviderPresets } from "../../providers/derive";
 import { providerCodexAccountMode } from "../../providers/registry";
 import { routedSlug, slugEquals } from "../../providers/slug-codec";
 import { clearAccountQuotaCache, clearProviderQuotaCache, fetchProviderAccountQuotas, fetchProviderQuotaReports, supportsPerAccountQuota } from "../../providers/quota";
@@ -407,20 +407,25 @@ export async function handleOauthAccountRoutes(ctx: ManagementContext): Promise<
       }
       stickyLimit = parsed;
     }
+    let googleProviderConfig: OcxProviderConfig | undefined;
+    if (provider === "google-antigravity" && body.enabled === false) {
+      googleProviderConfig = config.providers[provider] ?? deriveOAuthProviderConfig(provider);
+      if (!googleProviderConfig) {
+        return jsonResponse({ error: "google-antigravity OAuth provider is unavailable" }, 500);
+      }
+    }
     setOAuthPoolConfig(config, provider, {
       enabled,
       autoSwitchThreshold: threshold,
       ...(strategy !== undefined ? { strategy } : {}),
       ...(stickyLimit !== undefined ? { stickyLimit } : {}),
     });
-    if (provider === "google-antigravity" && body.enabled === false) {
-      const providerConfig = config.providers?.[provider];
-      if (providerConfig) {
-        providerConfig.oauthAccountFailover = {
-          ...providerConfig.oauthAccountFailover,
-          enabled: false,
-        };
-      }
+    if (googleProviderConfig) {
+      config.providers[provider] = googleProviderConfig;
+      googleProviderConfig.oauthAccountFailover = {
+        ...googleProviderConfig.oauthAccountFailover,
+        enabled: false,
+      };
     }
     saveConfigPreservingClaudeCode(config);
     reconcileLiveStateStores();
diff --git a/tests/account-pool-management-api.test.ts b/tests/account-pool-management-api.test.ts
index 642db1d6d8..dbae8701f0 100644
--- a/tests/account-pool-management-api.test.ts
+++ b/tests/account-pool-management-api.test.ts
@@ -4,7 +4,7 @@ import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { handleCodexAuthAPI } from "../src/codex/auth-api";
-import { saveConfig } from "../src/config";
+import { readConfigDiagnostics, saveConfig } from "../src/config";
 import { startServer } from "../src/server";
 import { handleOauthAccountRoutes } from "../src/server/management/oauth-account-routes";
 import type { OcxConfig } from "../src/types";
@@ -468,6 +468,32 @@ describe("Anthropic account pool strategy management API", () => {
     expect(isGenericOAuthFailoverEnabled(config, "google-antigravity")).toBe(false);
   });
 
+  test("turning the Google pool off creates and persists a missing provider opt-out", async () => {
+    const config = baseConfig();
+    delete config.providers["google-antigravity"];
+    saveConfig(config);
+    const req = new Request("http://localhost/api/oauth/accounts/pool", {
+      method: "PUT",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ provider: "google-antigravity", enabled: false }),
+    });
+    const response = await handleOauthAccountRoutes({
+      req,
+      url: new URL(req.url),
+      config,
+      deps: {},
+      convergeCodexCatalog: async () => ({ status: "unchanged" }),
+      syncClaudeAgentDefsBestEffort: async () => {},
+    });
+
+    expect(response?.status).toBe(200);
+    expect(config.providers["google-antigravity"]?.oauthAccountFailover).toEqual({ enabled: false });
+    expect(isGenericOAuthFailoverEnabled(config, "google-antigravity")).toBe(false);
+    const reloaded = readConfigDiagnostics().config;
+    expect(reloaded.providers["google-antigravity"]?.oauthAccountFailover).toEqual({ enabled: false });
+    expect(isGenericOAuthFailoverEnabled(reloaded, "google-antigravity")).toBe(false);
+  });
+
   test("updating a disabled Google pool without an enabled field preserves generic failover intent", async () => {
     const config = baseConfig();
     config.googleAntigravityAccountPool = { enabled: false };
diff --git a/tests/google-antigravity-docs-contract.test.ts b/tests/google-antigravity-docs-contract.test.ts
index 38d8ef6e8a..3c484ac160 100644
--- a/tests/google-antigravity-docs-contract.test.ts
+++ b/tests/google-antigravity-docs-contract.test.ts
@@ -2,6 +2,28 @@ import { describe, expect, test } from "bun:test";
 
 const LOCALES = ["", "fr", "ja", "ko", "ru", "tr", "zh-cn", "zh-tw"] as const;
 
+const DIRECT_DISABLE_MARKER: Record<(typeof LOCALES)[number], string> = {
+  "": "without changing the separate generic OAuth fallback",
+  fr: "sans modifier le basculement OAuth générique distinct",
+  ja: "別個の汎用 OAuth fallback は変更されません",
+  ko: "별도의 일반 OAuth fallback은 변경되지 않습니다",
+  ru: "отдельный generic OAuth fallback не изменится",
+  tr: "ayrı genel OAuth fallback ayarını değiştirmez",
+  "zh-cn": "不会改变独立的通用 OAuth fallback",
+  "zh-tw": "不會改變獨立的通用 OAuth fallback",
+};
+
+const PARTIAL_UPDATE_MARKER: Record<(typeof LOCALES)[number], string> = {
+  "": "updates that omit `enabled` preserve",
+  fr: "omet `enabled` conserve",
+  ja: "`enabled` を省略した部分更新は",
+  ko: "`enabled`를 생략한 부분 업데이트는",
+  ru: "Частичное обновление без `enabled` сохраняет",
+  tr: "`enabled` alanını atlayan kısmi güncellemeler",
+  "zh-cn": "省略 `enabled` 的 部分更新会保留",
+  "zh-tw": "省略 `enabled` 的 部分更新會保留",
+};
+
 function providersReferencePath(locale: string): string {
   const prefix = locale ? `${locale}/` : "";
   return `../docs-site/src/content/docs/${prefix}reference/configuration/providers.md`;
@@ -33,4 +55,24 @@ describe("Google Antigravity documentation contract", () => {
     )).text();
     expect(text).toContain("per-account probe (Anthropic and Google Antigravity today)");
   });
+
+  test("every provider reference distinguishes direct disable from operational off", async () => {
+    for (const locale of LOCALES) {
+      const text = await Bun.file(new URL(providersReferencePath(locale), import.meta.url)).text();
+      const section = headingSection(text, "### `googleAntigravityAccountPool`");
+      const compactSection = section.replace(/\s+/g, " ");
+      expect(section, providersReferencePath(locale)).toContain(
+        "ocx account auto-switch google-antigravity off",
+      );
+      expect(section, providersReferencePath(locale)).toContain("PUT/PATCH /api/oauth/accounts/pool");
+      expect(section, providersReferencePath(locale)).toContain("enabled: false");
+      expect(section, providersReferencePath(locale)).toContain(
+        "providers.google-antigravity.oauthAccountFailover.enabled",
+      );
+      expect(section, providersReferencePath(locale)).toContain("`true`");
+      expect(compactSection, providersReferencePath(locale)).toContain(DIRECT_DISABLE_MARKER[locale]);
+      expect(compactSection, providersReferencePath(locale)).toContain(PARTIAL_UPDATE_MARKER[locale]);
+      expect(section, providersReferencePath(locale)).not.toContain("ocx account pool google-antigravity");
+    }
+  });
 });