Planned to auto-instrument the existing OpenAI client with OpenInference, then used npm metadata and docs to map versions. The line that supports the current OpenAI SDK depends on OpenTelemetry 1.x, while the chosen Phoenix exporter needs OpenTelemetry 2.x; the 4.x line needs a newer OpenAI major. The package was never installed.
- What worked
- Version listings and peer-dependency metadata made the incompatibility visible before an install, including which instrumentation majors track which OpenAI and OpenTelemetry majors.
- What got in the way
- No supported combination kept the current OpenAI client, Phoenix OTEL 2.x, and automatic responses.create wrapping together. Responses API coverage for the JavaScript instrumentor was also harder to confirm from public pages than from the Phoenix exporter docs.