OpenInference

OpenInference is an open-source set of conventions and instrumentation plugins built by Arize AI, designed to complement OpenTelemetry for tracing AI and LLM applications. It is natively supported by Arize Phoenix but works with any OpenTelemetry-compatible backend. The project is hosted on GitHub under the Apache License 2.0 and is actively maintained with contributions from the community.

What It Is

OpenInference provides a specification and a growing library of auto-instrumentation packages that capture telemetry from LLM invocations, vector store retrievals, agent tool calls, and surrounding application context. The specification is transport- and file-format-agnostic, designed to work alongside JSON, ProtoBuf, and DataFrames. It extends OpenTelemetry's general-purpose tracing model with AI-specific semantic conventions, making it possible to observe what happens inside LLM calls, RAG pipelines, and multi-agent workflows without building custom tracing from scratch.

Instrumentation Coverage

OpenInference ships instrumentation packages across Python, JavaScript, Java, and Go. The Python library alone covers a wide range of popular frameworks and SDKs:

• LLM providers: OpenAI, Anthropic, MistralAI, AWS Bedrock, Google GenAI, Google ADK, VertexAI, Groq
• Agent frameworks: LangChain, LlamaIndex, DSPy, CrewAI, Haystack, Autogen AgentChat, PydanticAI, smolagents, Agno, BeeAI, Pipecat, Strands Agents, OpenAI Agents SDK, Claude Agent SDK
• Utilities: LiteLLM, Instructor, Portkey, Guardrails, MCP

JavaScript packages cover OpenAI, Anthropic, LangChain.js, AWS Bedrock, BeeAI, MCP, Vercel AI SDK, TanStack AI, and Claude Agent SDK. Java support includes LangChain4j, Spring AI, and annotation-based manual tracing via ByteBuddy. Go support covers OpenAI and Anthropic SDKs.

Span Processors and Interoperability

Beyond direct instrumentation, OpenInference includes span processors that normalize and convert data from other observability libraries — specifically OpenLIT and OpenLLMetry (Traceloop) — into the OpenInference format. This allows teams already using those libraries to route their traces into Phoenix or Arize AX without re-instrumenting their code.

Supported Destinations

Traces captured with OpenInference can be sent to:

• Arize Phoenix (open-source LLM observability platform)
• Arize AX (Arize's managed observability service)
• Any OTEL-compatible collector

Update: Recent Release Activity

The repository shows active development as of mid-2026, with the latest release being python-openinference-instrumentation-strands-agents v0.1.3 published on June 11, 2026. The project has accumulated over 1,000 GitHub stars and 259 forks since its creation in December 2023. The GitHub topics list includes recent additions such as pydantic-ai, smolagents, mcp, openai-agents, and vercel, reflecting the project's rapid expansion to cover emerging agent frameworks and tooling.