All open source
Open SourceAnalytics6k

Arize Phoenix

Open-source AI observability and evaluation for LLM apps and agents.

Phoenix is Arize's open-source observability tool for experimentation, evaluation, and troubleshooting of AI and LLM applications. Built on OpenTelemetry and OpenInference, it captures traces from popular frameworks, runs LLM-as-a-judge evaluations, and visualizes embeddings to surface drift and clusters of failures. It runs locally in a notebook, as a container, or self-hosted.

Repository

Arize-ai/phoenix

Language

Python

What you'd build with it

  • Instrument LangChain, LlamaIndex, or raw OpenAI calls with OpenTelemetry tracing
  • Score outputs with built-in LLM-as-a-judge evals for hallucination and relevance
  • Explore embeddings to detect retrieval drift and clusters of problematic queries

Tags

llm-observabilityevalstracingopentelemetry