Skip to content

copilot: run ug copilot on a Unity Catalog model location - #1256

Draft
xi-db wants to merge 14 commits into
mainfrom
xi-db/stack/copilot-model-location
Draft

copilot: run ug copilot on a Unity Catalog model location#1256
xi-db wants to merge 14 commits into
mainfrom
xi-db/stack/copilot-model-location

Conversation

@xi-db

@xi-db xi-db commented Oct 22, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator

🥞 Stacked PR

Use this link to review incremental changes.

What did you change, and why?

Change: ug copilot --model-location <catalog>.<schema> runs Copilot on the Unity Catalog model services of one schema, with them in /model. Claude Code and Codex already support model locations; Copilot only had system.ai and, in #1220, provider services.

  • Discovery, as for Claude Code.
    • The models come from the gateway's Anthropic model listing with the Databricks-Model-Service-Parent-Schema header, the same listing Claude Code's picker uses.
    • The listing names services by gateway alias (anthropic-aigw-<hash>-<catalog.schema.name>); Copilot registers and sends the full name, which the gateway routes as a model service.
    • The header only scopes the listing. Inference routes by the model id, so the registry carries no header.
  • Wire per service.
    • Services named like Claude or GPT models (…claude-sonnet-4-6, …gpt-5-6-luna) use the same wire as the system.ai catalog. GPT 5+ goes on Responses on /ai-gateway/mlflow/v1, because GPT 5+ rejects the temperature Copilot sends on the Anthropic wire.
    • Everything else (names that don't say which model backs them) goes on the Anthropic wire, as Claude Code sends every service. The gateway translates for other models.
    • claude-opus-4-1 is skipped, as in the system.ai catalog.
    • Claude limits are capped at 200K where Copilot's catalog limits exceed it, as for provider services, since a model service may not serve the 1M window.
  • Selection. --model-location, the saved "Custom Model Location" in ug configure and the managed unity_catalog_location now apply to Copilot like Claude Code and Codex (MODEL_LOCATION_TOOLS). --provider and --model-location stay mutually exclusive. The managed Databricks model list is ignored under a location.

How do you know it works?

Testing:

  • Unit:
    • location catalog: alias stripping, the name-based wire with the Anthropic fallback, claude-opus-4-1 skipped, the listing called with the schema, an empty schema;
    • a launch through the real proxy with a stubbed Copilot process: registry ids and wires, no header, 200K limits, start model;
    • ug copilot --model-location reaching the launch.
    • just lint is clean.
  • Integration (new tests/integration/test_ug_copilot_model_location.py, non-blocking Copilot lane): from fresh state, ug copilot --workspace … --model-location <model-discovery schema> --model <its Claude-compatible service> -p … completes a file task, with the answering model checked from Copilot's own session events.
  • Live (ug from this branch, a Databricks workspace):
    • --model-location system.ai started on system.ai.claude-sonnet-5-5 and answered.
    • --model system.ai.gpt-5-6-luna answered on dbx-responses. The same model on the Anthropic wire failed with "Unsupported parameter: 'temperature'", which is why GPT-named services use Responses.

Limits: a GPT 5+ model behind a service whose name doesn't say GPT goes on the Anthropic wire and can hit the same temperature rejection. Telling those apart needs each service's backing model, a possible follow-up.

Stacked on #1220.

This pull request and its description were written by Isaac.

@xi-db xi-db added the ug-review Run the automated UG review label Oct 22, 2026
@xi-db xi-db mentioned this pull request Oct 22, 2026
Open
@xi-db
xi-db force-pushed the xi-db/stack/copilot-model-location branch 5 times, most recently from 76d745b to 856a99d Compare October 23, 2026 10:00
xi-db and others added 14 commits October 23, 2026 10:24
@xi-db Isaac
copilot: route ug copilot through Model Provider Services
8cc4086 
ug copilot --provider <catalog>.<schema>.<name> registers the service's models in
the session registry: Claude models of Anthropic and Bedrock services on the
Anthropic endpoint, chat models of OpenAI and Azure OpenAI services on the
gateway's native OpenAI route (Responses for GPT 5+ and o-series, Chat
Completions otherwise). Every registry provider sends the
Databricks-Model-Provider-Service header, so the proxy needs no routing change.

Provider ids are classified whole (gpt-5.5, us.anthropic.claude-sonnet-4-6),
not as system.ai names, and dated or versioned Claude ids map to Copilot's
catalog id. A service that allows every model lists them through the gateway;
an empty listing (which is what a caller without permission gets) asks for
--model. Subscription-relay services are refused for Copilot. ug configure
offers matching services, and a managed model_provider_service applies.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: sort dotted provider versions newest first
aac0873 
Provider ids write versions with dots (gpt-5.6), which the version sort read as
unversioned, so a provider session started on its oldest GPT.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: fix provider model ids, allow-all start model and listing er…
e88532c 
…rors

- Region-less Bedrock and Mantle Claude ids (anthropic.claude-*) map to
  Copilot's catalog id, limits and family order again.
- An allow-all service starts only on --model or the managed default: the
  gateway lists a fixed catalog for it, not what the provider account serves.
  A --model in Copilot's dbx-<wire>/<id> form is accepted as the id.
- A failed allow-all listing reports its cause instead of being treated as an
  empty list.
- o-series models sort newest first; provider Claude limits are capped only
  where they exceed 200K.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: skip provider targets by their declared APIs, not by name
12e9b12 
A provider target lists its native_api_types. A target that declares none of
the chat APIs (Anthropic Messages, Chat Completions, Responses), such as an
embedding deployment, is skipped. The API types only exclude: the gateway
serves some models on APIs they don't declare (gpt-5.2-chat declares only Chat
Completions and answers on Responses). This replaces the name pattern; the
gateway listing for allow-all services is left unfiltered, as for Codex.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: list provider service models through a public databricks helper
4ba8436 
list_provider_service_models picks the Anthropic or Codex listing for a
provider type and returns (ids, reason), reporting an empty listing as empty
rather than a failure. Copilot no longer imports the private Codex catalog
helper or matches another module's error text to tell the two apart.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: start allow-all provider sessions on a listed model, like Co…
5c22ee9 
…dex and Claude

Without --model or a managed default, an allow-all service now starts on
Copilot's default pick from the gateway's listing instead of refusing to
launch, matching ug codex and ug claude. ug still asks for --model when the
listing is empty or fails.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: accept --workspace on ug copilot, like ug claude and ug codex
ae7a2c2 
The launch targets that workspace and sets it up first when it isn't
configured yet. Before, the flag was passed through to Copilot itself.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: reuse shared rules for provider services
b407499 
- One rule for what an agent can use a Model Provider Service for, shared by
  service_usable_for_tool and resolve_provider_service. A subscription relay
  is refused for every agent but Claude Code, not only Copilot.
- ANTHROPIC_API_PROVIDER_TYPES names the provider types the gateway serves
  on its Anthropic route, for both the listing and Copilot's classifier.
- Copilot classifies provider ids with the same name rules as system.ai
  models, and derives its chat API set from its wire routes.
- The Codex listing reuses catalog_slugs; the provider-service catalog is
  built once; Copilot's provider OpenAI base URL moves into copilot.py.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: one documented regex for bare model names; wire constants
e2598ad 
bare_model_name() in databricks.py replaces Copilot's split/removeprefix
chain with one regex, with examples of each id shape it handles. The wire
names Copilot writes to its registry are constants.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: one model classifier and one catalog builder
5eba8ed 
classify_model() holds every rule that maps a model id to Copilot's wire, for
Databricks model services and Model Provider Service targets alike, and
replaces classify_provider_model. build_catalog() builds every catalog from
(model, api_types) pairs and replaces build_provider_catalog. launch picks
its catalog through _session_catalog(). A system.ai o-series model now uses
Responses, like o-series provider targets.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
Name the AI Gateway API types in constants
b70393e 
The API types the gateway lists in supported_api_types and native_api_types
(anthropic/v1/messages, openai/v1/responses, ...) are shared constants, used
by Copilot's wire table and by Databricks endpoint discovery.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: say which provider is meant
abbb738 
Our new code used "provider" for a Model Provider Service, a Copilot
registry provider and the upstream vendor. Name each one: LAUNCH_PROVIDER_KEY
is LAUNCH_PROVIDER_SERVICE_KEY, provider_catalog is provider_service_catalog,
_without_provider_prefix is _without_registry_prefix, and comments, docstrings
and messages spell out which one they mean.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: drop reasoning_effort from Chat Completions requests
08da820 
Copilot sends reasoning_effort on Chat Completions. Only models that aren't GPT
5+ or o-series use that wire, and OpenAI and Azure reject the field for them
("Unrecognized request argument supplied: reasoning_effort"), so gpt-4o-mini
and gpt-4.1-mini failed through a Model Provider Service. The session proxy
now removes it from Chat Completions bodies; other requests pass unchanged.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db Isaac
copilot: run ug copilot on a Unity Catalog model location
a04cb29 
ug copilot --model-location <catalog>.<schema> registers the schema's model
services in the session registry, from the gateway's Anthropic model listing
as for Claude Code. Services are sent by full name; the gateway routes them
as model services. Services named like Claude or GPT models use their usual
wire (GPT 5+ on Responses, since GPT 5+ rejects the temperature Copilot sends
on the Anthropic wire); the rest use the Anthropic wire, which the gateway
translates for other models. The saved Custom Model Location and the managed
unity_catalog_location apply to Copilot like Claude Code and Codex.

Co-authored-by: Isaac <no-reply@databricks.com>
@xi-db
xi-db force-pushed the xi-db/stack/copilot-model-location branch from 856a99d to a04cb29 Compare October 23, 2026 10:27
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

ug-review Run the automated UG review

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@xi-db