Files
zed/crates/language_models
pixelrick 5255d9c721 Do not probe LM Studio and llama.cpp at startup when disable_ai is true (#66842)
## Summary

When `disable_ai: true` is set, Zed still sent model-discovery HTTP
requests to local providers at startup: constructing
`LmStudioLanguageModelProvider` and `LlamaCppLanguageModelProvider`
kicked off an eager fetch to `http://localhost:1234/api/v0/models` and
`http://localhost:8080/v1/models` respectively, before any AI UI was
opened. #66029 gated credential reads behind `ai_disabled`, but these
discovery fetches read no credentials, so the probes kept firing.

This gates `restart_fetch_models_task` in both providers on the existing
`ai_disabled` helper, so no discovery request is made while AI is
disabled. The gate covers the eager construction-time fetch, restarts on
settings changes, and `refresh_models`, and clearing `fetch_model_task`
cancels any fetch that is in flight when AI gets disabled mid-run. Once
`disable_ai` is set to false, discovery resumes through the existing
flows (opening the agent panel or the provider's settings page).

Closes #66792

## Testing

- Added a regression test per provider
(`does_not_fetch_models_when_ai_is_disabled`): with `disable_ai: true`,
constructing the provider results in zero HTTP requests; after
re-enabling AI, restarting discovery fetches models normally.
- `cargo test -p language_models` - 132 passed, 0 failed (Linux).
- `./script/clippy -p language_models` - clean.

To verify manually: put a listener on the probed ports (`nc -lv 1234`
and `nc -lv 8080`), launch Zed with `"disable_ai": true` in settings,
and observe no connection is attempted at startup. Then set `disable_ai:
false`, open the agent panel, and discovery connects.

## Self-review

- [x] I've reviewed my diff for quality, security, reliability, and
performance.
- [x] UI changes follow the
[checklist](https://zed.dev/docs/development/ui-checklist). (This pr
makes no UI changes.)
- [x] Tests cover the new or changed behavior.

Release Notes:

- Fixed LM Studio and llama.cpp model discovery requests being sent at
startup despite `disable_ai` being true
2026-11-11 17:22:55 +00:00
..