Self-host arize-phoenix
@arizeai/phoenix-cli@1.22.0AI tracing, LLM evals, and prompt ops for agents
docker run -p 6006:6006 -p 4317:4317 -i -t arizephoenix/phoenix:latest- Stars
- 12 258
- Ports
- 6006 (Web UI plus the OTLP HTTP trace collector endpoint /v1/traces), 4317 (OTLP gRPC trace collector)
arizephoenix/phoenixWhat is arize-phoenix?
Arize Phoenix is an AI observability and evaluation platform for teams building LLM applications and agents. A single Phoenix server collects traces over OpenTelemetry OTLP on HTTP and gRPC and presents them in a web UI, and the same workspace adds LLM-as-a-judge, code-based and human-reviewed scoring, versioned datasets, experiments that rerun the same inputs against new application versions, prompt versioning with a playground and span replay, an embedded AI engineering agent called PXI, and an MCP endpoint that lets coding agents query traces, datasets, and experiments. Instrumentation covers a wide range of LLM providers, agent frameworks and languages, and Python plus TypeScript clients talk to its REST API.
You can run it locally with pip or conda (pip install arize-phoenix, then phoenix serve), with uvx for no install at all, from the arizephoenix/phoenix container image using Docker or Docker Compose, or on Kubernetes with Kustomize or the bundled Helm chart; one-click templates also exist for Railway, Render, Google Cloud Run, Azure Container Instances, and AWS CloudFormation. Storage defaults to a SQLite file, while production setups point Phoenix at PostgreSQL 14 or newer through an environment variable. Authentication ships disabled; switching it on adds user accounts with admin, member and viewer roles, system and user API keys, secrets storage, password-reset email over SMTP, OAuth2/OIDC single sign-on, LDAP directories, login lockout and an optional stricter password policy.
Phoenix is distributed under the Elastic License 2.0, and its docs state that running it on your own infrastructure or cloud account is free with no gated features. Basic web analytics are collected unless telemetry is turned off, and one setting suppresses every outbound request for air-gapped installs. Arize separately sells a managed enterprise product, Arize AX, which the docs position alongside Phoenix on those same instrumentation standards.
arize-phoenix alternatives
LangfuseLangfuse is a self-hosted, open-source LLM engineering platform for tracing, prompt management, evaluations, and analytics — used by 50,000+ companies.CompareLangfuse with arize-phoenix
heliconeHelicone is a self-hosted, open-source LLM observability platform and AI gateway for 100+ models with cost tracking, agent tracing, prompt management, and a playground.Comparehelicone with arize-phoenix
TraceloopOpen-source observability for your GenAI or LLM application, based on OpenTelemetryCompareTraceloop with arize-phoenix
laminarOpen-source observability platform for AI agents. Alternative to LangSmith with tracing, Signals failure detection, evals, dashboards, MCP/CLI access.Comparelaminar with arize-phoenix
openlitOpen-source platform to trace, evaluate and guard AI agents with OpenTelemetry: evals, prompt hub, cost and GPU monitoring, self-hosted via Docker Compose or Helm.Compareopenlit with arize-phoenix
hearthSelf-hosted observability dashboard for home AI clusters monitoring vLLM, llama.cpp, SGLang, Ollama, LiteLLM with real metrics and training observability.Comparehearth with arize-phoenix
ObservalSelf-hosted, open-source registry for coding agent Skills, MCP servers, hooks and Agents, with session traces, AI insights and review governance for dev teams.CompareObserval with arize-phoenix
phenixcodePhenixCode lets developers query codebases with semantic search and AI chat, running on local models or cloud APIs, with no subscription fees.Comparephenixcode with arize-phoenix
All self-hosted alternatives to Braintrust → All self-hosted alternatives to LangSmith → All self-hosted alternatives to Langfuse →
Should you self-host arize-phoenix?
arize-phoenix scores 75/100 on selfhost.directory (how scoring works) and ranks #1 of 2 in LLMOps. It has 12 258 GitHub stars, +496 in the last 30 days.
It shipped 160 stable releases in the past year. The latest, @arizeai/phoenix-cli@1.22.0, came out on 11 November 2026. New versions land roughly every 1 days.
Setup is easy: most people get it running in one sitting. An official Docker image is available. It is Elastic-2.0-licensed and written mostly in Python.
Its closest alternative on selfhost.directory is Langfuse (85/100, 37 353 stars). It scores higher overall, so compare the two before you pick. See arize-phoenix vs Langfuse side by side.
Features
- Built-In MCP ServerDescribed as beta in the docs
- Data retention policies
- Dataset management
- Experiments
- Human annotations
- LLM-as-a-judge evaluation
- LLM tracing
- Multi-user accounts
- OpenTelemetry OTLP ingestion
- Prompt Management
- Prompt Playground
- Role-based Access Control
- Secrets Management
- Span replay
arize-phoenix README
🇺🇸 English • 🇨🇳 中文 • 🇹🇼 繁體中文 • 🇯🇵 日本語 • 🇵🇹 Português • 🇧🇷 Português • 🇰🇷 한국어 • 🇪🇸 Español • 🇩🇪 Deutsch • 🇫🇷 Français • 🇮🇱 עברית • 🇸🇦 العربية • 🇷🇺 Русский • 🇵🇱 Polski • 🇨🇿 Čeština • 🇳🇱 Nederlands • 🇹🇷 Türkçe • 🇺🇦 Українська • 🇻🇳 Tiếng Việt • 🇵🇭 Tagalog • 🇮🇩 Indonesia • 🇹🇭 ไทย • 🇮🇳 हिन्दी • 🇧🇩 বাংলা • 🇵🇰 اردو • 🇷🇴 Română • 🇸🇪 Svenska • 🇮🇹 Italiano • 🇬🇷 Ελληνικά • 🇭🇺 Magyar • 🇫🇮 Suomi • 🇩🇰 Dansk • 🇳🇴 Norsk
FAQ
Which SMTP service should I use?
The docs recommend a reputable transactional email provider, suggesting services such as Resend, Mailgun, Sendgrid or Postmark, or one from your cloud vendor.
How do I configure automatic trace retention and cleanup?
Set PHOENIX_DEFAULT_RETENTION_POLICY_DAYS to the number of days traces should be kept (0 keeps everything); cleanup runs weekly and individual projects can override the default in the UI.
How do I configure Phoenix for air-gapped deployments?
Set PHOENIX_ALLOW_EXTERNAL_RESOURCES=false so Phoenix makes no outbound internet requests of its own, disabling Google Fonts, telemetry, PXI lookups, the WebAssembly sandbox download and update checks. For the offline WASM sandbox, point PHOENIX_WASM_BINARY_PATH at a local copy of the runtime.
More in LLMOps
bishengAn Apache-2.0, self-hosted LLM devops platform for enterprise AI: GenAI workflows, RAG knowledge bases, agents, model management, evaluation, SFT and OCR.
FastChatSelf-hosted, open-source LLM serving platform: run open chat models behind a Gradio web UI, a CLI, or an OpenAI-compatible REST API.- openlitOpen-source platform to trace, evaluate and guard AI agents with OpenTelemetry: evals, prompt hub, cost and GPU monitoring, self-hosted via Docker Compose or Helm.
SuperAGIOpen-source, self-hosted framework for building, deploying and running autonomous AI agents, with a web GUI, action console, toolkits and multiple vector databases.- TraceloopOpen-source observability for your GenAI or LLM application, based on OpenTelemetry
0 comments