Arize Phoenix
An open-source AI observability and evaluation platform that traces model calls, retrievals, tool use and custom logic to debug agents. Built on OpenTelemetry and the OpenInference spec, runnable locally, self-hosted, on Kubernetes or via managed cloud.
- Website
- https://arize.com/phoenix/
- Who it's for
- Developers who want local, self-hostable observability and evals without shipping data to a SaaS platform.
- Pricing
- Open-source core is free and self-hosted. Managed cloud starts free with paid tiers whose prices are not published.
- Category
- Agent operations
- Last verified
- 2026-10-22 (current)
- Confidence
- high
Key facts
- Open source and self-hostable
- Accepts traces over OpenTelemetry (OTLP) using the OpenInference spec
- Auto-instrumentation for LlamaIndex, LangChain, DSPy, Mastra and the Vercel AI SDK
- Supports tracing, evals, prompt iteration and experiments
- Deploys locally, via Docker or Helm/Kubernetes
Sources
Every claim above is drawn from these; we do not publish unsourced specifications.
Notes & uncertainties
Managed cloud pricing is not published; treated as not stated.
Something changed or out of date? Tell us — corrections are how this stays worth reading.