Summary
Split from #58.
trace/contrib/openai traces /v1/chat/completions and /v1/responses, but it does not route /v1/embeddings. Calls made through the official openai-go client via client.Embeddings.New() therefore fall through without emitting Braintrust spans.
What is missing
- Route
/v1/embeddings in trace/contrib/openai/traceopenai.go
- Add an embeddings tracer that records model, input, usage/token metrics, and response metadata as a Braintrust
llm span
- Add tests covering embeddings path matching and traced embeddings requests
Upstream surface
Local repo files inspected
trace/contrib/openai/traceopenai.go
trace/internal/utils_test.go
Summary
Split from #58.
trace/contrib/openaitraces/v1/chat/completionsand/v1/responses, but it does not route/v1/embeddings. Calls made through the officialopenai-goclient viaclient.Embeddings.New()therefore fall through without emitting Braintrust spans.What is missing
/v1/embeddingsintrace/contrib/openai/traceopenai.gollmspanUpstream surface
github.com/openai/openai-goclient.Embeddings.New()Local repo files inspected
trace/contrib/openai/traceopenai.gotrace/internal/utils_test.go