Arize Phoenix
OpenTelemetry-based tracing, evals, datasets and prompt playground for LLM apps, run on your own machine or servers.
Self-hosted tracing, evals, datasets and a prompt playground built on OpenTelemetry.
Arize Phoenix is Arize's open-source AI observability platform for experimentation, evaluation and troubleshooting. It suits developers who want LangSmith-style traces and evals running locally or in their own cluster, as long as they are not planning to resell it as a service.
What it is
Phoenix records your LLM application's runtime as OpenTelemetry traces, then lets you build versioned datasets, run experiments on prompt, model or retrieval changes, and score outputs with LLM-based evals. It also includes a prompt playground that can replay traced LLM calls, prompt version control with tagging, and a built-in agent called PXI for debugging traces.
Who it's for
It fits Python, TypeScript, Java or Go teams already using frameworks like LangChain, LlamaIndex, OpenAI Agents SDK or Vercel AI SDK that want to see and score what their app does. It does not suit anyone who wants to host it for paying customers, because the licence forbids that, or teams that want a managed service from the open-source project itself.
Key features
OpenTelemetry-based tracing
Instrumentation follows OpenTelemetry and the OpenInference conventions, so traces are not locked to one vendor.
Versioned datasets and experiments
Build datasets of examples and compare changes to prompts, LLMs and retrieval against them.
Replay traced calls in playground
Open a traced LLM call in the playground, change the prompt or model parameters and run it again.
Prompt version control
Manage prompt changes with version history, tagging and experiments.
PXI built-in agent
An AI engineering agent inside Phoenix helps debug traces and iterate on prompts.
Remote MCP server
Connect Claude Code, Cursor and other MCP clients to your instance's /mcp endpoint to query traces, datasets and experiments.
Up and running in 3 steps
Our honest verdict
What's great
- The README lists instrumentation packages for a long run of frameworks (OpenAI Agents, LlamaIndex, DSPy, CrewAI, LangGraph, Pydantic AI and more) across Python, JavaScript, Java and Go.
- One command, phoenix serve after pip install arize-phoenix, gives you a working instance on a laptop, with uvx as a no-install option.
- Traces use OpenTelemetry, and span processors convert data from OpenLIT and OpenLLMetry traces.
- Separate lightweight client, OTel and evals packages exist for Python and TypeScript, plus a CLI, so you can talk to a deployed Phoenix without installing the whole platform.
Where it falls short
- The Elastic License 2.0 is not OSI-approved: you may not offer Phoenix to third parties as a hosted or managed service, and you may not remove its licence-key functionality.
- For production you run it yourself: the README points to Docker images and a Helm chart for Kubernetes, so you carry the hosting, upgrades and storage.
- The managed option is a different product, Arize AX, with its own pricing; the open-source project itself has no hosted version.
- The README warns that the Azure one-click template serves plain HTTP and the Google Cloud button builds from source in Cloud Shell, so those routes need extra work before they suit production.
Pricing & license
Phoenix itself is free to run under the Elastic License 2.0; your cost is the machine or cluster you host it on. The managed alternative is Arize AX (arize.com/pricing): AX Free at $0 with 25k spans a month, 1 GB ingestion and 15-day retention; AX Pro at $50 a month with 50k spans, 10 GB and 30 days; AX Enterprise on custom pricing, the only tier listed with self-hosted deployment.
Open remote AI-training roles
767 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Is Arize Phoenix open source?
The source is public but the licence is Elastic License 2.0, which is source-available rather than OSI-approved. You can use and modify it, but you cannot provide it to others as a hosted or managed service.
Does Phoenix only work with LangChain?
No. Its README says it is vendor, language and framework agnostic and lists instrumentation for OpenAI, Anthropic, LlamaIndex, DSPy, CrewAI, LangGraph and many others.
Is there a hosted Phoenix?
The README says Phoenix is open-source and self-hosted, and points to Arize AX as the managed platform using the same instrumentation.
