Phoenix

Phoenix is Arize AI’s open-source platform for AI observability and evaluation, built to help teams experiment with, troubleshoot, and systematically improve LLM applications. It gives you deep visibility into what your application actually did at runtime, and the tooling to measure whether changes make it better or worse.

Overview

Phoenix is built on OpenTelemetry and the OpenInference standard, so it is vendor, language, and framework agnostic, with ready-made instrumentation for OpenAI, Anthropic, LangChain, LangGraph, LlamaIndex, CrewAI, the Vercel AI SDK, and many more. You can run it locally with a single pip install, in Docker, or on Kubernetes, and it includes an MCP server so coding agents like Claude Code and Cursor can query your traces directly. The project is very actively maintained, with updates in October 2026 and more than 11,900 GitHub stars. For teams that outgrow self-hosting, Arize offers Arize AX, a managed platform built on the same instrumentation.

Key Features

• OpenTelemetry-based tracing of LLM calls, retrievers, tools, and agents
• LLM-assisted evaluations for relevance, toxicity, hallucination, and more
• Versioned datasets and experiments to track prompt and model changes
• Playground for comparing models and replaying traced calls
• Prompt management with versioning, tagging, and experimentation
• Built-in MCP server and an in-product AI debugging assistant

💳 Pricing

Phoenix itself is free and open source for self-hosting. Arize AX, the company’s managed production platform with enterprise features, is a paid product.

Who Is It For

AI engineers and platform teams who need to debug, evaluate, and continuously improve production LLM and agent applications.