AX Free
$0USD / per month
A free tier for builders, with no credit card required, running on the SaaS deployment.
- 10 Signal issues per month
- 25k trace spans per month
- 1 GB ingestion volume per month
- 15-day retention
- Unlimited users and evaluations
AI Code Review & Testing Tools
- Introduction
- Open-source, local-first platform for tracing, evaluating, experimenting on, and iterating prompts for AI agents and LLM applications.
- Added on
- 11/6/2026
- Pricing
- Freemium
- Official links
- Website, company & support ↗
PRODUCT OVERVIEW
What is Arize Phoenix?
Arize Phoenix is an open-source, local-first observability and evaluation platform for AI agents and LLM applications. It records traces of LLM requests as they move through multi-step workflows, using the OpenInference specification and OpenTelemetry-compliant telemetry to show where retrieval, tool execution, or reasoning broke down. Teams can run online and offline evaluations, including LLM-as-a-judge scoring for relevance, toxicity, and response quality, and can experiment with datasets, playgrounds, and prompts. Phoenix runs locally or self-hosted in your own environment, and can be started with a single command or connected to an app through the Phoenix CLI. It is aimed at AI engineers and developers who need to debug and improve production AI systems, with an optional path to the managed Arize AX platform.
How to use Arize Phoenix
- 1
Start a local Phoenix server
Run the Phoenix server locally, for example with the uvx arize-phoenix serve command, and leave it running at the local address shown.
- 2
Connect your application
Use the Phoenix CLI via npx to walk through connecting an application to Phoenix, then hand off to your coding agent.
- 3
Add tracing
Instrument your LLM application so requests are recorded as traces, following the OpenInference specification and OpenTelemetry-compliant telemetry.
- 4
Run evaluations
Apply online evals on traces and spans or offline evals on datasets and experiments, including LLM-as-a-judge scoring.
- 5
Experiment and iterate on prompts
Use playgrounds, datasets, experiments, and prompt iteration to compare and improve prompts before shipping.
Arize Phoenix Features
Open-source, local-first observability and evaluation platform
Trace recording for multi-step LLM and agent workflows
OpenInference specification and OpenTelemetry-compliant telemetry
Online evals on traces and spans, and offline evals on datasets and experiments
LLM-as-a-judge evaluation of relevance, toxicity, and response quality
Playgrounds, datasets, experiments, and prompt iteration
Runs locally or self-hosted in your own environment
Phoenix CLI setup that hands off to a coding agent
Arize Phoenix Use Cases
Debugging where an LLM workflow broke during retrieval or tool execution
Evaluating agent response quality before releasing to production
Comparing prompts and experiments across datasets
Tracing multi-step agent requests to find root causes of poor responses
Self-hosting observability inside your own environment
Arize Phoenix FAQs
Yes. Arize Phoenix is described as an open-source, local-first platform for tracing, evaluation, experimentation, and prompt iteration.
Yes. Phoenix can run locally or be self-hosted in your environment, and the docs include self-hosting deployment options such as terminal, Docker, and Kubernetes.
The Phoenix page shows the command uvx arize-phoenix serve to get started locally, which starts a local Phoenix server.
Phoenix records LLM traces so you can follow requests through multi-step workflows and see where retrieval, tool execution, or reasoning went wrong.
Phoenix supports online evals on traces and spans and offline evals on datasets and experiments, including LLM-as-a-judge scoring for relevance, toxicity, and response quality.
Yes. Phoenix includes playgrounds, datasets, experiments, and prompt iteration so teams can compare and improve prompts.
The managed Arize AX platform has an AX Free tier with no credit card required, offering 10 Signal issues, 25k trace spans, and 1 GB ingestion per month with 15-day retention.
Phoenix is the open-source, local-first platform; Arize AX is the managed AI engineering platform, and the pricing page presents moving to AX as an option when you need it.
Arize Phoenix Official Information
- Company
- Arize AI, Inc
- Website
- https://arize.com/phoenix
- Documentation
- https://arize.com/docs/phoenix
- Contact
- https://arize.com/contact/
Arize Phoenix Pricing
Phoenix itself is an open-source, local-first platform. The listed plans are for the managed Arize AX platform, which the pricing page presents as the next step when you need it. AX Free requires no credit card.
AX Pro
$50USD / per month
A paid tier for AI-native teams, running on the SaaS deployment with higher limits than AX Free.
- 25 Signal issues per month
- 50k trace spans per month
- 10 GB ingestion volume per month
- 30-day retention
- Unlimited users and evaluations
AX Enterprise
Price to be verified
A custom plan for teams with scaled AI use cases, available as SaaS or self-hosted deployment.
- Unlimited Signal issues
- Custom trace span volume and ingestion
- Custom retention
- SaaS or self-hosted deployment
- Unlimited users and evaluations
COMMUNITY REVIEWS
Arize Phoenix User Reviews
No reviews yet.
