Summary

The WebSocket responses handler rejected bare model strings (e.g. gpt-4o) on the /openai/v1/responses integration path with a 400 "failed to parse model string" error. This PR fixes the two ParseModelString call sites so the integration path infers openai as the default provider from the request URL, matching the behavior of the HTTP path.

Changes

• Added inferDefaultProviderFromPath(path string) schemas.ModelProvider helper in wsresponses.go. Returns schemas.OpenAI for paths starting with /openai/, and an empty provider for all other paths (including the unified /v1/responses, which is multi-provider by design and intentionally requires an explicit provider/model prefix).
• Threaded a defaultProvider schemas.ModelProvider parameter from handleUpgrade (where ctx.Path() is available) through eventLoop and handleResponseCreate to convertEventToRequest. Both ParseModelString call sites (lines 177 and 422 on origin/main) now receive the correct default.
• Design choice: path-sniff approach over storing a field on session or config, because the fix is entirely local to wsresponses.go, requires no new types, and reads cleanly at the call site. The session object is not the right place for request-scoped routing data. The parameter threading is shallow (3 levels).

Type of change

• Bug fix

Affected areas

• Transports (HTTP)

How to test

Run the handler unit tests, which cover the new helper and the parse behavior on both paths:

cd transports
go build ./bifrost-http/handlers/...
go vet ./bifrost-http/handlers/...
go test -count=1 -cover ./bifrost-http/handlers/...

Expected output:

ok  github.com/maximhq/bifrost/transports/bifrost-http/handlers  ~1.2s  coverage: 8.2% of statements

Live reproduction (requires a running Bifrost instance with an OpenAI provider configured and allow_direct_keys: true):

# Before fix: returns 400 "failed to parse model string"
printf '{"type":"response.create","model":"gpt-4o","input":[{"type":"message","role":"user","content":[{"type":"input_text","text":"hi"}]}]}\n' | \
  websocat -n1 "ws://localhost:8080/openai/v1/responses" \
  -H='Authorization: Bearer sk-your-key'

# After fix: reaches the upstream (401 on a fake key, 200 on a real key)

Screenshots/Recordings

N/A

Breaking changes

• No

Related issues

Closes #2995. Extracted from #2775.

Security considerations

The path-sniff only changes which default provider value is passed to ParseModelString. It does not affect auth, key selection, or any other security-relevant logic. The unified /v1/responses path is unchanged.

Checklist

• I read docs/contributing/README.md and followed the guidelines
• I added/updated tests where appropriate
• I updated documentation where needed
• I verified builds succeed (Go and UI)
• I verified the CI pipeline passes locally if applicable