Arize Phoenix
Arize Phoenix is a self-hosted platform for observability and evaluation of LLM applications and AI agents. It records runs via OpenTelemetry and assesses answers and retrieval with datasets and experiments.
- Project
- Arize Phoenix
- Publisher
- Arize AI
- Repository
- github.com/Arize-ai/phoenix
- Documentation
- arize.com/docs/phoenix
- Languages
- Python, TypeScript
- Category
- Testing and evaluating AI agents
What is Arize Phoenix?
Phoenix is Arize AI’s source-available observability platform for experimentation, evaluation and troubleshooting. It runs locally, in containers or in the cloud and is vendor- and language-agnostic. It is licensed under the Elastic License 2.0; for managed production use, Arize additionally offers the commercial Arize AX.
For tracing, Phoenix uses OpenInference instrumentation based on OpenTelemetry. According to the README, integrations exist for the OpenAI Agents SDK, Claude Agent SDK, LangGraph, Vercel AI SDK, Mastra, CrewAI, LlamaIndex and DSPy, among others.
Core capabilities
- Tracing
- Recording the runtime of an LLM application: following model calls, tool calls and retrieval steps.
- Evaluation
- Checking and comparing answers and retrieval with LLM-based evaluations.
- Datasets and experiments
- Creating versioned example sets and measuring the effect of changes to prompts, models or retrieval.
- Playground and prompt management
- Optimising prompts, comparing models, replaying recorded calls and managing prompt changes under version control.
- MCP server
- An MCP endpoint lets coding agents query traces, datasets and experiments.
When Phoenix fits
Phoenix fits when you want not only to build agents but to understand their behaviour and measure changes before rolling them out – and keep that data in your own infrastructure. Because the instrumentation is based on open standards, it can be combined with different frameworks.
Using Phoenix effectively in an AI-first business model
Evaluation is not a one-off sign-off. An agent is never finished: data, models and goals change, and with them the quality of results. A tool like Phoenix makes that change visible – provided there are criteria derived from the business goal.
Define which cases an agent must handle reliably, at what deviation someone intervenes and who decides. Datasets and experiments then become part of operation, not just development. Because traces can contain customer data, storage location and access rights belong in this planning as well.
From tool to outcome
In AI-first coaching you clarify which outcomes AI agents should be accountable for in your go-to-market and within which framework. REVAGO then designs, builds and runs your system. You pay for results – not for activity.
Frequently asked questions
What is Arize Phoenix?
Arize Phoenix is a self-hosted platform by Arize AI for observability and evaluation of LLM applications and AI agents. It offers OpenTelemetry-based tracing, evaluation, versioned datasets, experiments, a playground and prompt management.
Is Arize Phoenix free?
Phoenix is source-available and self-hosted and is licensed under the Elastic License 2.0. For managed operation, Arize offers the commercial product Arize AX, which uses the same instrumentation.
Which frameworks does Phoenix work with?
Phoenix is vendor- and language-agnostic. According to the README, integrations exist for the OpenAI Agents SDK, Claude Agent SDK, LangGraph, Vercel AI SDK, Mastra, CrewAI, LlamaIndex and DSPy, among others, as well as for various model providers.
Further reading in the wiki
A neutral overview, not a recommendation of individual providers. REVAGO is not affiliated with the listed projects. Descriptions are based on the projects’ public information; features and licences may change.