Arize Phoenix overview and key facts (October 2026)

Arize Phoenix is an open-source observability and evaluation tool for LLM applications. Developers trace model calls, score outputs and compare prompt or model versions on their own infrastructure.

Phoenix captures traces through OpenTelemetry and OpenInference instrumentation and integrates with LangChain, LlamaIndex, CrewAI, LangGraph and major model providers. Evaluations use LLM judges, code checks or human labels, and datasets and experiments compare application versions. A playground, prompt management, a built-in debugging agent and a remote MCP server round out the tool. Phoenix runs locally, in Docker or on Kubernetes. It is maintained by Arize AI, Inc. under the Elastic License 2.0 and had about 11,600 stars with commits on September 28, 2026.

• Traces what an agent didEach model call, retrieval step and tool use is recorded as a trace. Developers see where an answer went wrong instead of guessing from the final output.
• Scores outputs with LLM judgesEvaluators rate responses for criteria such as relevance or hallucination, with an LLM, code or human labels. Teams monitor quality across many traces rather than reading samples.
• Tests changes before releaseDatasets and experiments rerun the same examples against a new prompt or model. Differences in quality show up before users are affected.

• Built on open standards OpenTelemetry and OpenInference, so traces are not tied to one vendor.
• Self-hosted Phoenix is free; the managed Arize AX platform is available as an upgrade path.

Phoenix is an open-source project used worldwide that can be run in any self-managed environment.

Check the Elastic License 2.0, which allows internal use but restricts offering Phoenix as a managed service to third parties. Traces can contain user prompts and personal data, so define retention and access. Self-hosting means your team operates the database and updates.

No reviews yet. Be the first to share how your team uses it.

Every review shows its verification level. We read each one before it goes live. How we check reviews