Buying snapshot
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Decision point | What Arize states |
|---|---|
Licence | Elastic License 2.0 (ELv2) |
Installation | pip or conda: |
Deployment | Self-hosted: on a local machine, as a Docker container (images on Docker Hub), or on Kubernetes with a Helm chart |
Instrumentation | OpenTelemetry-based, with OpenInference auto-instrumentation for frameworks such as OpenAI Agents SDK, Claude Agent SDK, LangGraph, LlamaIndex, CrewAI and DSPy |
Languages | Python and TypeScript client packages; instrumentation packages for Python, JavaScript, Java and Go |
Data and telemetry | Self-hosted traces stay on your infrastructure. By default Phoenix collects basic web analytics (page views, UI interactions), not trace data or evaluation results; opt out with PHOENIX_TELEMETRY_ENABLED=false |
Our take
Choose Phoenix if you want to trace and evaluate LLM or agent applications on your own machine or servers, and your company accepts the ELv2 licence. If you would rather not run the service yourself, look at Arize AX, the same vendor's managed platform.
Evaluate Arize Phoenix
Comparisons
Where it stands out
Fast local start. Arize says Phoenix runs locally on a laptop in under a minute, so a developer can inspect agent traces before any platform decision.
One loop from trace to fix. Phoenix traces every step (prompts, retrievals, tool calls, outputs), lets people or an LLM-as-judge label what worked and what broke, turns traces into datasets, and runs experiments to check a change.
Open instrumentation. Arize describes Phoenix as vendor agnostic, working with any model, framework or language, with native OpenTelemetry support and no proprietary lock-in.
Built for coding agents. A remote MCP server at the instance's
/mcpendpoint lets Claude Code, Cursor and other MCP clients query traces, datasets and experiments.npx @arizeai/phoenix-cli setupdetects the framework and LLM provider and installs the matching instrumentation. PXI, a built-in agent, helps debug traces and iterate on prompts.Prompt management. Prompt changes can be managed and tested with version control, tagging and experiments.
Product screenshot
Trade-offs to weigh
Check the licence first. Arize calls Phoenix open source and licenses it under ELv2. If your company approves software by licence, check ELv2 against that list before adoption.
You secure what you host. Phoenix is self-hosted, so production hardening is your job. For example, Arize notes its Azure template serves plain HTTP and should sit behind a TLS proxy, such as an Application Gateway, before production use.
TypeScript evals are early. The
@arizeai/phoenix-evalsTypeScript evaluation library is marked as an alpha release.
Pricing and plan conditions
Phoenix itself is self-hosted under ELv2, so what you pay for is the infrastructure you run it on. The plans below are for Arize AX, Arize's managed platform. Arize's own advice is to stay local with Phoenix and move to AX only when you need it. AX Free and AX Pro run as SaaS only; a self-hosted AX deployment needs AX Enterprise. Arize also offers startup pricing, which teams apply for.
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Plan | Published price | What it includes |
|---|---|---|
AX Free | Free, no credit card required | 25k trace spans per month, 1 GB ingestion per month, 15-day retention, 10 Signal issues per month; SaaS; unlimited users and evals; community troubleshooting support |
AX Pro | $50 per month | 50k trace spans per month, 10 GB ingestion per month, 30-day retention, 25 Signal issues per month; SaaS; unlimited users and evals; email troubleshooting support; standard SLAs |
AX Enterprise | Custom | Custom trace spans, ingestion volume and retention; unlimited Signal issues; SaaS or self-hosted; unlimited users and evals; dedicated troubleshooting support; custom SLAs |
Public signals
About these signals
Verified factual signals from public sources. They indicate observable activity or interest, not total adoption, product quality, or cost.
1.1k GitHub commits 90d12.2k GitHub stars0 vulnerabilities across 2 packages
See all signals from 5 sources
npm · @arizeai/phoenix-client@7.18.2 · PyPI · arize-phoenix@21.0.0
Alternatives
Arize AX is the managed option from the same vendor. Arize says the same OpenTelemetry/OpenInference instrumentation works with both Phoenix and AX, so moving between them does not mean re-instrumenting the application. Arize positions AX around Signal, which surfaces issues, finds root causes and generates review-ready fixes, plus Managed Agents and monitoring across a whole agent swarm. Choose AX when you want Arize to host the service; stay on Phoenix when traces must stay on your own infrastructure and Enterprise terms are not needed.
Alternatives to Arize Phoenix
The reviewed substitutes for Arize Phoenix among the LLM observability & evaluation, and what would make each one the better answer.
Direct alternatives
Reviewed substitutes: products bought for the same job, where a team picks one.
- Braintrust
- Both score LLM output against datasets and record production traces. They are alternatives for the same evaluation workflow.
- Helicone
- Both are open LLM observability tools a team runs itself. Helicone integrates as a proxy and Phoenix through OpenTelemetry, but the purchase decision is one or the other.
- Langfuse
- Both are LLM tracing and evaluation platforms built on OpenTelemetry and run by the team itself. A team picks one as its trace destination.