lsm / neokai / 26796026000

82%

Build Type

push

github

Committed by web-flow

Commit Message

feat: auto-fetch model list from custom endpoints via /v1/models (#2055)

* feat: auto-fetch model list from custom endpoints via /v1/models

Add backend RPC handler `customEndpoints.listModels` that probes
OpenAI-compatible `/v1/models` and Ollama `/api/tags` endpoints,
normalizes responses, and caches results for 30s.

Update custom endpoint editor UI with a "Fetch models" button that
calls the RPC and presents returned models as a selectable list.
Users can check models and add them in bulk, or continue typing
manually. The picker shows model count and last-fetched timestamp.

Works in AddProviderModal, CustomEndpointsSettings, and
ProvidersSettings.

* fix: address review feedback for custom endpoint model fetching

- URL construction: strip /v1 and /v1/models suffixes before appending
  to avoid double paths like /v1/v1/models
- Anthropic normalization: accept type: 'model' and display_name fields
- Anthropic auth: send x-api-key + anthropic-version alongside Authorization
- OpenAI object filter: only reject entries with an explicitly incompatible
  object type, accept entries that omit the field entirely
- Clear fetched models when editor opens/closes or connection fields change
- Extract duplicated handleFetchModels into shared useFetchModels hook
- Update test mocks and add backend tests for new behavior

* fix: address second round of review feedback

- Strip /chat/completions from baseUrl before appending /v1/models
- Add activeRequestRef stale-response guard in useFetchModels
- Use 15s RPC timeout for listCustomEndpointModels
- Surface parseHeaders errors instead of swallowing silently
- Clear selectedFetchedModelIds when fetchedModels changes
- Add backend test for /chat/completions stripping

* fix: reset loading state and strip more bridge URL suffixes

- Reset fetchingModels=false when invalidating stale fetch in useFetchModels
- Strip /api/chat from ollama-native baseUrl before appending /api/tags
- Strip /v1/messages and /v1/messages/... (continued)

Coverage Stats

9896 of 14446 branches covered (68.5%)

Branch coverage included in aggregate %.

164 of 199 new or added lines in 6 files covered. (82.41%)

180 existing lines in 5 files now uncovered.

82926 of 97571 relevant lines covered (84.99%)

297.27 hits per line