Arize Phoenix
Arize Phoenix is an open-source AI observability platform for experimentation, evaluation, and troubleshooting of LLM applications. It provides tracing, evaluation, datasets, experiments, and prompt management capabilities. It is designed for developers and AI engineers building and debugging LLM applications.
✨ Key features
- OpenTelemetry-based tracing for LLM applications
- LLM-powered evaluation of response and retrieval quality
- Versioned datasets for experimentation and fine-tuning
- Track and evaluate prompt, LLM, and retrieval changes
- Playground for prompt optimization and model comparison
- Remote MCP server for querying traces and datasets
🎯 Use cases
- Trace and debug LLM application runtime behavior
- Benchmark application performance with LLM evals
- Manage and test prompt changes systematically
- Compare models and parameters in the playground
- Connect coding agents to query traces and datasets
📦 Installation
🧰 Requirements: Python 3.8+ or Node.js for subpackages; no API keys required for self-hosted Phoenix.
pip install arize-phoenix
phoenix serve
Or run with no install using uvx:
uvx arize-phoenix serve
🚀 Usage
phoenix servenpx @arizeai/phoenix-cli setup⚠️ Good to know
Phoenix is open-source and self-hosted; for managed production workflows, Arize offers Arize AX.
❓ FAQ
How do I install Phoenix?
Install via pip with pip install arize-phoenix and run phoenix serve. Alternatively, use uvx arize-phoenix serve without installation.
What frameworks are supported for tracing?
Phoenix supports popular frameworks like OpenAI Agents SDK, LangGraph, LlamaIndex, CrewAI, and many more via OpenInference instrumentation.
Can I deploy Phoenix in the cloud?
Yes, Phoenix can run locally, in containers, or in the cloud. Docker images and a Helm chart are available for Kubernetes deployment.
What is the difference between Phoenix and Arize AX?
Phoenix is open-source and self-hosted, while Arize AX is a managed platform. Both use the same OpenTelemetry/OpenInference instrumentation.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.