Arize Phoenix
What is Arize Phoenix?
Arize Phoenix is an open-source agent tracing and evaluation platform for AI engineers that shows each step an agent takes, then turns traces into datasets, experiments, and scored evaluations. It includes Tracing, Evaluation, Iteration, and Agent Integrations, and supports OpenTelemetry plus LlamaIndex. Teams use it with GitHub, Microsoft, and Atlassian, and can run it locally, in Docker, on Kubernetes, or in the cloud.
Last verifiedHow we evaluate
At a glance
- Phoenix is best for AI engineers who need trace-level visibility and repeatable evaluation for agents.
What it does
Phoenix traces every step of an LLM or agent call — prompts, retrievals, tool calls, outputs — and layers evaluation, dataset/experiment tooling, and a prompt-engineering IDE on top. It ships as a Python/TypeScript package (uvx arize-phoenix serve starts a local instance in one command) or a Docker/Helm/Kubernetes deployment. Instrumentation is built on OpenTelemetry plus OpenInference, a semantic-convention spec Arize authored and open-sourced for representing LLM/agent spans; OpenInference is now consumed by third parties outside Arize's own stack (e.g., Envoy AI Gateway), which is a reasonable signal it has become a real interoperability layer rather than a vendor lock-in trick.
Self-hosting is free with no feature gates
Arize's docs state plainly that self-hosted Phoenix has no license fees and no feature gates — tracing, evals, datasets, experiments, and prompt management are all included, and data never leaves your infrastructure (air-gapped deployment is supported). This is a genuine strength: many competitors that call themselves 'open source' still meter self-hosted usage or gate features behind a paid tier once you're running your own instance. Deployment options include a terminal one-liner, Docker, Kubernetes/Helm, and one-click templates for AWS CloudFormation, Railway, Render, Google Cloud Run, and Azure.
Licence: source-available, not OSI open source
Phoenix is released under the Elastic License 2.0 (ELv2), confirmed directly in the GitHub repo's LICENSE file. ELv2 permits free use, self-hosting, and modification, but it explicitly prohibits offering the software to third parties as a hosted or managed service that exposes substantial Phoenix functionality — i.e., you can't stand up a competing 'Phoenix-as-a-service.' Arize's own docs describe Phoenix as 'fully open-source,' which is the vendor's framing; ELv2 is not on the OSI's approved list, so buyers who need a strictly OSI-licensed dependency (for compliance or redistribution reasons) should check this before assuming 'open source' means what it usually means.
Hosted option: Arize AX pricing
If you don't want to self-host, Arize sells the same tracing/eval product as a managed SaaS called Arize AX, on a separate pricing page (arize.com/pricing, verified 2026-08-27). Free tier: 10 issues/month, 25k spans/month, 1GB ingestion/month, 15-day retention, unlimited users and evals. Pro tier: $50/month for 25 issues/month, 50k spans/month, 10GB ingestion/month, 30-day retention. Enterprise: custom pricing, custom limits, SaaS or self-hosted deployment with Arize support. These limits are modest — 25k-50k spans/month is easy to exceed with even light agent traffic — so teams expecting to scale past a hobby project should model out ingestion volume against the Enterprise tier rather than assume Free/Pro will hold.
Maturity and activity
The Arize-ai/phoenix GitHub repo has 11,205 stars and 1,080 forks (verified via GitHub API, 2026-08-27), was created November 2022, and had its most recent push within the last 24 hours of this check — an actively maintained project, not an abandoned one. It has 890 open issues, which is high in absolute terms but consistent with a widely-used, actively-developed tool rather than a sign of neglect given the commit cadence. The PyPI package arize-phoenix is at version 20.4.0, and GitHub releases show a steady stream of component releases (core package, evals, client, sqlean plugin) rather than a single monolithic version bump — indicative of a modular, still-evolving codebase.
Company backing
Phoenix is built and maintained by Arize AI, which raised a $70M Series C in February 2025 led by Adams Street Partners with participation from Microsoft's M12, per Arize's own press release and corroborated by third-party funding trackers (The SaaS News, TechNews180). This is vendor-disclosed funding, not independently audited, but multiple outlets report the same figure and lead investor, which is reasonable corroboration for a private funding round. A venture-backed company behind a free OSS tool cuts both ways: it means ongoing investment in Phoenix, but Arize's business model depends on converting self-hosters to the paid Arize AX product, which is worth knowing when reading their comparison pages.
Security and compliance
Arize publishes a real trust center (arize.com/trust-center) listing SOC 2 Type II, ISO/IEC 27001, PCI DSS 4.0, GDPR, and HIPAA compliance claims, plus a named CISO and a public bug bounty program run through Yogosha. A public status page (status.arize.com) tracks Application, Ingestion, API, and Alyx components, all reporting operational as of this check. These apply to the hosted Arize AX platform; self-hosted Phoenix's security posture is whatever you configure on your own infrastructure, which the docs are explicit about (Phoenix support that includes formal security backing is only bundled with the paid on-premise Arize AX option, not the free self-hosted OSS package).
Frequently asked questions
What is Arize Phoenix?
Arize Phoenix is an open-source agent tracing and evaluation platform for AI engineers that shows each step an agent takes, then turns traces into datasets, experiments, and scored evaluations. It includes Tracing, Evaluation, Iteration, and Agent Integrations, and supports OpenTelemetry plus LlamaIndex. Teams use it with GitHub, Microsoft, and Atlassian, and can run it locally, in Docker, on Kubernetes, or in the cloud.
What is Arize Phoenix used for? Who is it for?
Arize Phoenix is used for Tracing, Evaluation, and Iteration. It's built for AI engineers, ML teams, and Product teams building LLM features that need evidence-based iteration.
Does Arize Phoenix have an API and what does it integrate with?
Arize Phoenix doesn't publish a public API. It integrates with LlamaIndex.
Editor's read
Check whether your team needs self-hosting or local deployment before adopting Phoenix. The listing shows those options, so confirm your infrastructure can support them and that your workflow benefits from tracing, evaluation, and iteration on private agent data.
