Integrations Overview

Phoenix integrates with popular LLM frameworks and providers through OpenTelemetry-based instrumentation

Phoenix is built on OpenTelemetry and provides auto-instrumentation for popular LLM frameworks, SDKs, and providers. Tracing integrations allow you to observe your LLM application’s runtime behavior with minimal code changes.

Supported Frameworks

Phoenix provides first-class support for the most popular LLM application frameworks:

LangChain

Auto-instrument LangChain applications in Python and JavaScript

LlamaIndex

Trace LlamaIndex queries, retrievals, and agent workflows

OpenAI

Monitor OpenAI API calls including GPT-4, embeddings, and agents

Anthropic

Instrument Claude API calls with streaming support

OpenTelemetry

Custom instrumentation using OpenTelemetry primitives

LLM Providers

Phoenix supports instrumentation for all major LLM providers:

Python Integrations

Provider	Package	Description
OpenAI	openinference-instrumentation-openai	GPT models, embeddings, and function calling
Anthropic	openinference-instrumentation-anthropic	Claude models with streaming support
Google GenAI	openinference-instrumentation-google-genai	Gemini and PaLM models
AWS Bedrock	openinference-instrumentation-bedrock	Amazon Bedrock models
MistralAI	openinference-instrumentation-mistralai	Mistral models
VertexAI	openinference-instrumentation-vertexai	Google Vertex AI models
Groq	openinference-instrumentation-groq	Groq inference engine
LiteLLM	openinference-instrumentation-litellm	Unified interface for 100+ LLMs

JavaScript/TypeScript Integrations

Provider	Package	Description
OpenAI	@arizeai/openinference-instrumentation-openai	OpenAI Node.js SDK
LangChain.js	@arizeai/openinference-instrumentation-langchain	LangChain JavaScript framework
Vercel AI SDK	@arizeai/openinference-vercel	Vercel AI SDK streaming
BeeAI	@arizeai/openinference-instrumentation-beeai	BeeAI agent framework

Agent Frameworks

Phoenix supports popular agent and workflow frameworks:

• DSPy - openinference-instrumentation-dspy
• CrewAI - openinference-instrumentation-crewai
• Haystack - openinference-instrumentation-haystack
• Guardrails - openinference-instrumentation-guardrails
• Instructor - openinference-instrumentation-instructor
• Agno - openinference-instrumentation-agno
• Pydantic AI - openinference-instrumentation-pydantic-ai

Getting Started

Most integrations follow a simple three-step pattern:

1. Install the instrumentation package

   pip install openinference-instrumentation-{provider}
   
2. Register Phoenix tracer

   from phoenix.otel import register
   
   tracer_provider = register(
     project_name="my-llm-app",
     auto_instrument=True  # Auto-instrument based on installed packages
   )
   
3. Run your application

   Your LLM calls will now be automatically traced and sent to Phoenix.

How It Works

Phoenix integrations use OpenTelemetry to capture traces from your LLM applications:

1. Auto-instrumentation: Instrumentation packages automatically patch SDK methods to capture trace data
2. OpenInference conventions: Traces follow standardized semantic conventions for LLM observability
3. OTLP export: Trace data is exported to Phoenix using the OpenTelemetry Protocol (OTLP)
4. Zero-code changes: Most integrations require no changes to your application code

All Phoenix integrations are open source and part of the OpenInference project.

Next Steps

OpenAI Integration

Get started with OpenAI tracing

LangChain Integration

Instrument LangChain applications

Custom Instrumentation

Build custom traces with OpenTelemetry

All Integrations

Browse all available integrations