[Programcode]

Cost fix 4 of 7. Expensive models use up the daily AI allowance faster, and each user has a $2 weekly AI spend cap that leaves out Claude billed to our own Anthropic key (BYOK).

Not for merge yet: Adam has to approve the migrations below first.

What changes for users

• An Opus request counts as 5 of the day's requests, GPT-6.1 Sol as 3, Sonnet and DeepSeek V4 Pro as 2, and every other model as 1. Auto Pro counts as 5 because the quota is reserved before Auto picks a model. Auto Fast counts as 1.
• If a request doesn't fit in what's left today, the dialog says "This model counts as 5 uses, and you have 4 left today. Pick a lighter model or wait until midnight UTC." A lighter model still runs.
• If a model counts for more than the user's whole daily limit (only possible with a hand-set daily_limit_override below 5), the message says "This model counts as 5 uses, more than your daily limit of 3. Pick a lighter model." It no longer tells them to wait.
• Once a user's AI spend for the current week reaches $2, every AI request is refused with a "Weekly AI Allowance Used" dialog. It has an OK button and no upgrade offer. The week starts Monday 00:00 UTC.
• Refusal text now comes from the en/es/hu catalog everywhere, including pages that print error.message directly. Spanish and Hungarian users no longer see the server's English reason.

What counts toward the $2

There's a new column, ai_requests.cap_cost_usd, holding the part of a request's cost that counts:

• OpenRouter calls count OpenRouter's usage.cost, which is what OpenRouter charged against our credits.
  • A BYOK Claude call (is_byok: true) adds only OpenRouter's fee, usually $0, because Anthropic bills our own key.
  • A Claude call that falls back to OpenRouter credits (is_byok: false) has its full price in usage.cost, so it counts in full.
• Non-OpenRouter charges count in full: ElevenLabs voice now, and Exa research once Log the real cost of Anthropic (BYOK) calls and Exa research #179 lands (any usage record without a separate cap figure counts its full reported cost).
• Partly priced requests: when part of a request's cost is unknown, the known part counts. The cap is a lower bound, not all-or-nothing.
• Old rows don't count: a row with no known cap spend, or logged before this column existed, has NULL and is never counted. So Log the real cost of Anthropic (BYOK) calls and Exa research #179's backfill, which only rewrites cost_usd on old rows, can't push anyone toward the cap.
• Side effect: spend from earlier in the deploy week doesn't count. Cost is logged after the provider answers, so a user can go past the cap by about one request.

I chose this over matching the model name (as Adam asked). Research helper calls and Exa are merged into the parent request's row, so the row's model column doesn't say where each charge came from, but OpenRouter's own credit charge does.

Where to change the numbers

Everything lives in supabase/functions/_shared/usage-limits.ts:

• WEEKLY_AI_BUDGET_USD = 2 (null turns the cap off)
• MODEL_QUOTA_UNITS
• the trial and Pro daily defaults, 5 and 50

Changing a number needs an ai and ai-stream deploy, not a migration.

Migrations (additive: no DROP, no data rewrite)

1. 20261022130000_index_ai_requests_user_created.sql: CREATE INDEX CONCURRENTLY IF NOT EXISTS idx_ai_requests_user_created ON public.ai_requests (user_id, created_at); It's in its own file because of CONCURRENTLY.
2. 20261022130001_weighted_quota_weekly_budget.sql:
   • ALTER TABLE public.ai_requests ADD COLUMN IF NOT EXISTS cap_cost_usd numeric; The column is nullable with no default, so existing rows aren't rewritten. It also adds a column comment.
   • CREATE FUNCTION public.consume_ai_quota(p_user_id uuid, p_limit integer, p_units integer DEFAULT 1, p_weekly_cap_usd numeric DEFAULT NULL, p_quota_key text DEFAULT 'ai') RETURNS TABLE (allowed boolean, current_count integer, denied_by text). It's SECURITY DEFINER with search_path = public and raises unless auth.role() = 'service_role' and the arguments are sane (units ≥ 1, cap ≥ 0).
     • It sums cap_cost_usd since date_trunc('week', now() AT TIME ZONE 'UTC') AT TIME ZONE 'UTC' and refuses with denied_by = 'weekly' once the sum is ≥ the cap, without touching the counter.
     • It refuses with 'daily' when p_units > p_limit.
     • Otherwise it does one atomic INSERT … ON CONFLICT … DO UPDATE … WHERE request_count + p_units <= p_limit on daily_usage_counters, the same counter row and key (ai) the old function uses.
   • REVOKE ALL … FROM PUBLIC, anon, authenticated; GRANT EXECUTE … TO service_role.
   • consume_daily_quota is not touched. Import, share, voice and Crossfire signals keep calling it. Changing its arguments or result columns in place would have needed a DROP, so the weighted path got a new function instead.

Deploy order: apply the migrations first, then deploy ai and ai-stream.

• New functions on the old schema would fail closed with "Unable to verify your quota", and their cap_cost_usd insert would fail, so usage rows wouldn't be logged.
• Old functions on the new schema keep working unchanged.

Rollback: redeploy the previous ai and ai-stream. The new column and function can stay; nothing else uses them.

Verification (local, VPS; GitHub Actions is blocked by billing)

• Full npm run verify on 50498c12 passes: docs, lint, typecheck, 342 unit tests, 30 collaboration tests, 369 edge tests, build, test:sw, and 30/30 @smoke e2e.
• pgTAP: the shared DB gate lock was held by another workspace's stack, so I applied all migrations to a throwaway supabase/postgres:17.11.0.004 container instead (with storage stubs). ai_quota_weekly_budget_test.sql passes 17/17 and daily_usage_quota_test.sql 14/14. The other failures in that container were environment-only (no GoTrue columns; pgTAP in public). Rerun ./.superset/db-gate.sh --db-only before merge.
• New pgTAP file supabase/tests/ai_quota_weekly_budget_test.sql (17 tests). It covers:
  • grants and that consume_daily_quota still has one overload
  • weighted reservations, a request that exactly fills the limit, a request bigger than the limit, and zero units
  • that the old function shares the ai counter
  • the weekly cap ignoring BYOK rows (cost_usd only), NULL cap rows, last week's rows and other users' rows
  • a NULL cap turning the check off
  • that rows with only cost_usd, including backfilled ones, never count
  • that non-service callers are refused
• supabase/tests/daily_usage_quota_test.sql and scripts/test-daily-quota.sh are back to main's versions.
• Edge tests cover cap cost for BYOK (0), credit-billed Claude (full), merged calls, an Exa-style charge, a partly unknown request (known part only), voice, and the server's too-large message.
• Vitest covers translated refusal messages.
• The skill recipe recipes/usage-limits.spec.ts was updated to the weekly dialog and a cap_cost_usd: 2 row. I didn't run it in this round.

Reviews: Codex is rate-limited. The Opus review of 59d49c7 had one low finding, the wrong advice when a model can never fit, which is fixed here. Codex round 1 had one low finding, English refusals on es/hu screens, also fixed here.

Overlap

• Log the real cost of Anthropic (BYOK) calls and Exa research #179 changes normalizeOpenRouterUsage so reportedCostUsd = fee + upstream for BYOK. Whichever lands second must keep capCostUsd: usage.cost (the credit charge only) in that function. Exa usage needs no change: it counts through the reportedCostUsd fallback.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
debate-navigator	Ready	Preview	Oct 22, 2026 1:27pm UTC
debatelab-staging	Ready	Preview	Oct 22, 2026 1:27pm UTC

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[coderabbitai]

Warning

Review limit reached

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

Next included review available in 52 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

• Source mode: Auto
• Repository: Programcode/DebateLab
• Branch: cost/weighted-quota-budget
• File: .coderabbit.yaml
• Commit: 85aa5c1
• Layers: Repository YAML (base), Organization UI (inherited)
• Review profile: CHILL
• Plan: Free
• Run ID: 2abb475c-a4d5-4dfe-bb57-184cfa0777ca

📥 Commits

Reviewing files that changed from the base of the PR and between 50498c1 and 85aa5c1.

📒 Files selected for processing: 8

• docs/architecture.md
• src/lib/ai/error-display.ts
• src/locales/en.json
• src/locales/es.json
• src/locales/hu.json
• supabase/functions/ai-stream/index_test.ts
• supabase/functions/ai/handler-research.test.ts
• supabase/functions/ai/stream.test.ts

ℹ️ Recent review info

⚙️ Run configuration

• Source mode: Auto
• Repository: Programcode/DebateLab
• Branch: cost/weighted-quota-budget
• File: .coderabbit.yaml
• Commit: 50498c1
• Layers: Repository YAML (base), Organization UI (inherited)
• Review profile: CHILL
• Plan: Free
• Run ID: e21b47de-cc59-4565-9fdd-6b35ed862f2e

📥 Commits

Reviewing files that changed from the base of the PR and between 34fb585 and 50498c1.

📒 Files selected for processing: 39

• .claude/skills/verify-debatelab/features/README.md
• .claude/skills/verify-debatelab/features/usage-limits.md
• .claude/skills/verify-debatelab/recipes/usage-limits.spec.ts
• docs/architecture.md
• src/components/features/subscription/LimitReachedModal.tsx
• src/hooks/use-subscription-error.ts
• src/lib/ai/error-display.ts
• src/lib/ai/errors.ts
• src/lib/ai/limit-message.test.ts
• src/lib/ai/limit-message.ts
• src/lib/ai/request.ts
• src/locales/en.json
• src/locales/es.json
• src/locales/hu.json
• src/pages/AskPage.tsx
• src/pages/AttackPage.tsx
• src/pages/BrainstormPage.tsx
• src/pages/FactoidsPage.tsx
• src/pages/GeneratePage.tsx
• src/pages/SparringPage.tsx
• src/test/security-hardening.test.ts
• supabase/functions/_shared/cost.test.ts
• supabase/functions/_shared/cost.ts
• supabase/functions/_shared/entitlements.test.ts
• supabase/functions/_shared/entitlements.ts
• supabase/functions/_shared/usage-limits.ts
• supabase/functions/ai-stream/handler_test.ts
• supabase/functions/ai-stream/index.ts
• supabase/functions/ai-stream/index_test.ts
• supabase/functions/ai/debate-context.test.ts
• supabase/functions/ai/generate-honesty_test.ts
• supabase/functions/ai/handler-research.test.ts
• supabase/functions/ai/index.ts
• supabase/functions/ai/provider-errors_test.ts
• supabase/functions/ai/stream.test.ts
• supabase/functions/ai/voice.test.ts
• supabase/migrations/20261022130000_index_ai_requests_user_created.sql
• supabase/migrations/20261022130001_weighted_quota_weekly_budget.sql
• supabase/tests/ai_quota_weekly_budget_test.sql

Included review availability: This review used your included allowance. Your plan provides up to 1 included review per hour; 0 remain after this review.

📝 Walkthrough

The PR adds model-weighted daily AI quotas and a weekly AI spend cap. It records cap-cost amounts, enforces both limits in quota reservations, and returns limit details for localized client messages. Tests and documentation cover the quota rules and refusal outcomes.

Changes: AI usage limits

Layer / File(s)	Summary
Weekly spend accounting supabase/functions/_shared/cost.ts, supabase/functions/_shared/cost.test.ts, supabase/migrations/20261022130000_index_ai_requests_user_created.sql, supabase/migrations/20261022130001_weighted_quota_weekly_budget.sql	Cost results now carry capCostUsd. Merged usage retains known cap costs when total cost is unknown. The migration adds the nullable cap_cost_usd column and an index for user-created request lookups.
Quota reservation and enforcement supabase/functions/_shared/usage-limits.ts, supabase/functions/_shared/entitlements.ts, supabase/functions/_shared/entitlements.test.ts, supabase/migrations/20261022130001_weighted_quota_weekly_budget.sql, supabase/tests/ai_quota_weekly_budget_test.sql, supabase/functions/ai*/..., supabase/functions/ai-stream/..., src/test/security-hardening.test.ts	The consume_ai_quota function reserves weighted daily units and checks weekly spend. Entitlement checks pass model units and the weekly cap. AI endpoint responses include quota units, and tests cover denials, reservations, cost accounting, and updated quota RPC mocks.
Client limit messages src/lib/ai/limit-message.ts, src/lib/ai/limit-message.test.ts, src/lib/ai/errors.ts, src/lib/ai/request.ts, src/lib/ai/error-display.ts, src/hooks/use-subscription-error.ts, src/components/features/subscription/LimitReachedModal.tsx, src/pages/{Ask,Attack,Brainstorm,Factoids,Generate,Sparring}Page.tsx, src/locales/{en,es,hu}.json	Subscription errors now carry quota units and use localized messages for weighted daily limits and weekly-budget denials. The modal adds a weekly-budget variant, and AI pages pass quota units or display the localized inline message.
Usage-limit verification and documentation .claude/skills/verify-debatelab/recipes/usage-limits.spec.ts, .claude/skills/verify-debatelab/features/README.md, .claude/skills/verify-debatelab/features/usage-limits.md, docs/architecture.md	Playwright coverage checks weighted daily refusals, lighter-model requests, weekly-cap denials, provider calls, and counter values. The feature recipe and architecture documentation describe the limit rules and verification setup.

Estimated code review effort: 4 (Complex) | ~45 minutes

Sequence Diagram(s)

sequenceDiagram
  participant AIEndpoint
  participant checkRateLimit
  participant consume_ai_quota
  participant ai_requests
  participant AIProvider
  AIEndpoint->>checkRateLimit: Check request user and model
  checkRateLimit->>consume_ai_quota: Reserve units with daily limit and weekly cap
  consume_ai_quota->>ai_requests: Read current-week cap_cost_usd
  consume_ai_quota-->>checkRateLimit: Return allowance and denial reason
  checkRateLimit-->>AIEndpoint: Return limit result with units
  alt Request allowed
    AIEndpoint->>AIProvider: Send AI request
    AIProvider-->>AIEndpoint: Return provider response
    AIEndpoint->>ai_requests: Record usage with cap_cost_usd
  else Request denied
    AIEndpoint-->>AIEndpoint: Return rate-limit response
  end

Loading

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Note

🎁 Summarized by CodeRabbit Free

Your organization is on the Free plan. CodeRabbit will generate a high-level summary and a walkthrough for each pull request. For a comprehensive line-by-line review, please upgrade your subscription to CodeRabbit Essentials by visiting https://app.coderabbit.ai/settings/billing.

_{Comment @coderabbitai help to get the list of available commands.}

[vercel]

Deployment failed for project debatelab-staging with the following error:

Resource is limited - try again in 24 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/adam-gaborjanis-projects?upgradeToPro=build-rate-limit

[vercel]

Deployment failed for project debate-navigator with the following error:

Resource is limited - try again in 24 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/adam-gaborjanis-projects?upgradeToPro=build-rate-limit