openinference

Why we included this project

Tracing is how you find out what actually happened inside an LLM request, and OpenInference is one of the most complete open implementations of that for AI workloads. It combines a specification with a family of instrumentation plugins that emit standardized traces across Python, JS, Go, and Java SDKs, including OpenAI, LangChain, LlamaIndex, CrewAI, MCP servers, Bedrock, and Vertex. Because the same instrumentation fits so many stacks, you can feed every trace into one backend instead of hand-wiring a bespoke exporter per framework. It builds on OpenTelemetry rather than reinventing it, so traces stay transport- and backend-agnostic and work with Arize Phoenix, Arize AX, or any OTel-compatible store. Teams running mixed agent and RAG architectures will find the shared semantic conventions worth adopting even if they bring their own analysis tooling.