Arize Phoenix is an open-source observability and evaluation platform for LLM applications and agents. Teams instrument Python, TypeScript, and other runtimes with OpenTelemetry-based tracing, then inspect runs, score outputs, and iterate on datasets and prompts from a self-hosted UI.[1][2]
Quick verdict
Engineering teams that self-host LLM observability benefit most from Arize Phoenix: OTel-based tracing, built-in evaluators, versioned datasets and experiments, and a prompt playground cover the debug-to-iterate loop on operator infrastructure. The constraint is operational scope: free Elastic-2.0 self-hosting with air-gapped support, not OSI-approved licensing or managed SaaS, which sits in Arize AX. Teams wanting self-run tracing and evals should choose Phoenix; managed-tier comparison points to Langfuse.[1][2]
Key features
OpenTelemetry-based LLM tracing OpenTelemetry-based LLM tracing: capture hierarchical traces of model calls, tool use, and retrieval steps with auto-instrumentation for common frameworks.[1][2]
LLM evaluators for output quality LLM evaluators for output quality: score response and retrieval quality with built-in evaluators, human annotation queues, and tracked experiment runs.[1][2]
Versioned datasets and experiments Versioned datasets and experiments: version evaluation datasets and compare prompts and models side by side with repeatable experiment tracking.[1][2]
Prompt playground and versioning Prompt playground and versioning: replay traced calls across models and parameters, then version prompts with tags and test changes before rollout.[1][2]
Python and TypeScript clients plus APIs Python and TypeScript clients plus APIs: instrument applications with maintained SDKs and read traces and evals through REST and GraphQL interfaces.[1][2]
CLI and MCP access for agents CLI and MCP access for agents: let coding agents query traces and run evals from the terminal or through the MCP endpoint alongside the web UI.[1][2]
Local-run and self-hosted deploys Local-run and self-hosted deploys: start locally with pip or one command, then deploy with Docker, Compose, Kubernetes, Helm, or cloud templates, including air-gapped operation.[1][2]
Auth with SSO and access control Auth with SSO and access control: authentication is off by default and adds local accounts, API keys, role-based access, and OAuth2 identity providers when enabled.[1][2]
Who will love it
Teams debugging LLM apps in production: hierarchical traces, latency and cost context, and session inspection isolate failing tool and retrieval steps.[1][2]
Teams building eval loops: versioned datasets, built-in evaluators, and experiment comparison turn one-off prompt tweaks into repeatable benchmarks.[1][2]
Self-hosting teams with data constraints: free Elastic-2.0 self-hosting, air-gapped operation, and disablable telemetry keep traces on operator infrastructure.[1][2]
Who should avoid it
Teams wanting managed SaaS without operations: Phoenix itself is run by the adopter, so managed-service buyers should evaluate Arize AX instead.[1][2]
Teams expecting OSI-approved open source: Phoenix ships under Elastic License 2.0, a source-available custom license that bars offering the software as a hosted service.[1][2]
Non-engineering teams wanting turnkey AI: instrumentation, deployment, and auth setup still assume engineers who own code and infrastructure.[1][2]
Alternatives and compatibility
Langfuse is the natural catalogue comparison: both cover tracing, evals, datasets, and prompt iteration for LLM apps and agents.[1][2]
Phoenix fits teams prioritizing free self-hosted operation with OTel-native instrumentation and agent-accessible evals on their own infrastructure.[1][2]
Langfuse fits teams comparing managed-cloud tiers against self-hosting, including its usage-based Hobby, Core, Pro, and Enterprise options.[1][2]
Questions worth asking
What is the best Arize Phoenix alternative for non-technical users?
Is Arize Phoenix really no-code?
How much does Arize Phoenix cost per month?
Self-hosted Phoenix is free with no feature gates under the Elastic License 2.0, which bars offering the software itself as a hosted service. There is no per-seat monthly self-host price in the inspected official pages; vendor-backed managed and enterprise use is the separate Arize AX product, so confirm current AX pricing on arize.com before budgeting. Checked October 7, 2026.[1][2]
Can you self-host Arize Phoenix?
Sources and editorial review
Source research. Reviewed on October 7, 2026. Written by Sanjib Sah. AI assisted; reviewed by a human editor.
We have not run or tested this product.
- Arize Phoenix product page (official)
official · Observed October 7, 2026
Official product page: open-source positioning, tracing, evals, playground, local run and self-hosting, community scale, AX separation.
- Arize Phoenix GitHub repository (vendor-published)
official · Observed October 7, 2026
Vendor-published repository: scope, SDKs, deployment options, Elastic-2.0 license, auth model, community support channels.
Compared with these products
No explicit proprietary baseline is recorded.