Repository navigation
fix(voice): stop stale gateway warmups and false success logs - #764
Conversation
|
@vera review |
![protoreview[bot]](https://avatars.githubusercontent.com/in/4233969?s=60&v=4){kind=small}
There was a problem hiding this comment.
QA panel review — PASS
code-review · head 5a4ce762effe · formal
Clean pass: no findings from any panel member, and the verifier confirmed there was nothing to verify. No disagreements, no coverage gaps to flag. The PR is clear to merge on the review axis.
No findings — the review came back clean.
findings JSON (machine-readable)
[]
There was a problem hiding this comment.
QA panel review — PASS
code-review · head 5a4ce762effe · formal
Clean pass: no findings from any panel member, and the verifier confirmed there was nothing to verify. No disagreements, no coverage gaps to flag. The PR is clear to merge on the review axis.
No findings — the review came back clean.
findings JSON (machine-readable)
[]
There was a problem hiding this comment.
QA panel review — PASS
code-review · head 5a4ce762effe · formal
PR #764 is clean — no findings from any panel member, and the verifier confirmed there is nothing to verify. No disagreements, no coverage gaps to flag. Ship it.
No findings — the review came back clean.
findings JSON (machine-readable)
[]
1 participant
Problem and change
After selecting an OAuth-backed model in ORBIS, startup still sent an OpenAI chat-completions warmup to a stale environment-configured gateway. A rejected request then logged “LLM warm,” producing a misleading 401 alongside otherwise successful voice calls.
Warmup now resolves the active persona through the same model/key resolver as voice. Protocol-specific OAuth and in-process MLX providers skip this generic HTTP ping. Compatible endpoints use the saved URL, model, key and request options, and HTTP errors are no longer reported as success. Warmup remains optional and outside the native voice startup critical path.
Validation
.gitkeep.ruff check .,git diff --check, andbun run buildpassed.