Add first-class Ollama provider - #743
Conversation
TheGreatAxios
left a comment
There was a problem hiding this comment.
Critic lens · clean
The branch adds first-class keyless Ollama setup and dynamic /v1/models discovery through the existing OpenAI-compatible adapter.
No blocking correctness findings remain. Provider switching clears stale credentials in src/tui/provider-setup.ts:1593, and the submit boundary independently omits credentials for keyless Ollama in src/tui/provider-setup-submit.ts:56; regression tests cover both sequences.
TheGreatAxios
left a comment
There was a problem hiding this comment.
Primary review · clean
The branch adds a keyless Ollama provider with root-to-/v1 projection, dynamic model discovery, nonfatal recovery states, tests, and documentation.
No findings. origin/main...HEAD contains one scoped commit and 17 Ollama-related files; bun run check, git diff --check, and all GitHub checks pass. GitHub does not allow an author to approve their own PR, so this verdict is posted as a review comment.
Legacy Custom rows stored the OpenAI-compatible /v1 path. Treating that as a root threw while building backup sources and crashed session start for every provider. Strip /v1 once, keep true network failures as not-running, and seed the setup catalog so /model can switch among installed models.
A leftover Ollama Custom URL with an extra path threw while building sibling sources, so a healthy head provider could not start. Skip per-ref build failures in the backup list and still fail when the active provider cannot build.
015c5e1 to
930737a
Compare
Summary
/v1/modelsendpoint and handles unavailable, empty, and malformed responses without crashing.Verification
bun run checkpassesgit diff --checkpassesFixes CL-7330