Arize Phoenix provides tracing and observability for OpenAI's response APIs (completions, chat completions, embeddings) by auto-instrumenting your application. You can use the OpenInference OpenAI instrumentor to capture all OpenAI API calls and stream them to Phoenix for analysis and evaluation, including prompt/response pairs and associated metadata. This enables you to visualize, evaluate, and troubleshoot LLM responses in the Phoenix UI according to <https://docs.arize.com/phoenix/tracing/integrations-tracing/openai#P%3A4|Phoenix OpenAI integration docs> and <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/llm-traces-1.md|Phoenix LLM tracing tutorial>.
If you want to use the OpenAI "Responses API" specifically, Phoenix supports capturing and analyzing all response data via OpenTelemetry-based tracing, but does not provide a custom API for OpenAI's "Responses API" beyond this instrumentation. Would you like more detail on how to set up tracing for OpenAI responses in Phoenix?
Sources:
- <https://docs.arize.com/phoenix/tracing/integrations-tracing/openai#P%3A4|Phoenix OpenAI integration docs>
- <https://github.com/Arize-ai/phoenix/blob/main/docs/tracing/llm-traces-1.md|Phoenix LLM tracing tutorial>
