Arize Phoenix
Open-source LLM tracing and evaluation built on OpenTelemetry, with Arize AX as the managed enterprise option
Overview
Phoenix is Arize AI's open-source platform for tracing, evaluating and experimenting with LLM applications. It runs locally in a notebook, in a container or as a hosted instance, and collects traces through OpenInference, Arize's set of OpenTelemetry instrumentations for LangChain, LlamaIndex, the OpenAI and Anthropic SDKs, DSPy and many more. On top of traces, Phoenix offers LLM-as-judge evaluations, datasets, experiments for comparing prompts and models, and a prompt playground.
Phoenix is free to self-host with no feature gates. Teams that want a managed service, larger scale or enterprise controls can move to Arize AX, which shares the same instrumentation and adds monitoring, dashboards, SSO and SOC 2 compliance.
The Verdict
Who Should Use Phoenix?
Best For
- Teams that want free, self-hosted tracing and evals
- RAG debugging and retrieval evaluation
- OpenTelemetry-based stacks
- Data scientists who like notebook-first workflows
Not Ideal For
- Teams that want a fully managed tool without any setup (use Arize AX)
- Organizations needing an OSI-approved license (Phoenix uses Elastic License 2.0)
- Non-technical reviewers who need a simple annotation UI
What's Great
- Fully featured and free to self-host
- OpenTelemetry-native through OpenInference
- Strong built-in evaluators for RAG and agents
- Experiments and datasets for regression testing
- Smooth upgrade path to Arize AX
Watch Out For
- Self-hosting at scale means running your own database and storage
- Elastic License 2.0 restricts offering it as a hosted service
- Production monitoring and alerting are stronger in paid Arize AX
- UI can feel dense for newcomers
Pricing
View all features & details
Key Features
- OpenTelemetry tracing via OpenInference
- LLM-as-judge and code evaluators
- Datasets and experiments
- Prompt playground and prompt management
- Retrieval and embedding analysis
Integrations
- LangChain, LlamaIndex, DSPy, Haystack
- OpenAI, Anthropic, Bedrock, Vertex AI
- CrewAI, OpenAI Agents SDK, smolagents
- Python and TypeScript
How It Compares
| Feature | Phoenix | Langfuse | LangSmith | Braintrust |
|---|---|---|---|---|
| Self-host free | Yes | Yes | No | No |
| License | Elastic 2.0 | MIT | Proprietary | Proprietary |
| OpenTelemetry native | Yes | Accepts OTel | Partial | Partial |
| Evals and experiments | Yes | Yes | Yes | Core focus |
| Managed option | Arize AX | Langfuse Cloud | Yes | Yes |
Is this your tool? Add a badge