AI Potluck
Product / UX / Telemetry & observability

Phoenix

Arize AI

AI observability and evaluation toolkit from Arize AI, providing tracing, embeddings analysis, experiment tracking and LLM evaluation in a local-first workflow. It instruments applications through OpenTelemetry with integrations for LlamaIndex and LangChain, and deploys via Docker or a Helm chart.

The companion to the Arize AX platform, built by the same team and scored separately from it. Verified 2026-08-13 via the Arize-ai/phoenix repository and the Arize Phoenix product page.

Openness

2 high confidence
2.0
license
Elastic-License-2.0(ELv2, non-OSI: prohibits offering as a managed/hosted service to third parties)
source
public
commercial
Arize-AX-SaaS-is-separate-closed-product

The LICENSE file is the Elastic License 2.0 verbatim, which is not an OSI license, while Arize markets Phoenix as "the open source platform" - open-washing. The source is published and the platform is self-hostable, with Arize AX sold beside it as a separate closed SaaS product. The scale used for software has rungs at 1, 2, 4 and 5 and none at 3, and it places source you can read but not run freely at 2.

  • https://github.com/Arize-ai/phoenix/blob/main/LICENSE recorded 2026-08-13

    Elastic License 2.0 (ELv2) full text, read rather than assumed. It opens "Elastic License 2.0 (ELv2)" and its Limitations state "You may not provide the software to third parties as a hosted or managed service, where the service provides users with access to any substantial set of the features or functionality of the software." That is the non-OSI restriction the record scores on.

  • https://arize.com/phoenix/ recorded 2026-08-13

    The vendor page, which markets Phoenix as "The open-source platform for agent development and evaluation" while its own OSS Community block reads "ELv2 licensed. 10k+ GitHub stars." Self-hosting is a first-class path - "Self-Host ... your data on your own infrastructure" and a Helm/Kubernetes deployment option - so the published source is what you run.

Adoption

4 high confidence
4.0

arize-phoenix draws 2,313,184 PyPI downloads a month, inside the 1M-10M band. Broad OpenTelemetry-based framework integrations and roughly 11,000 GitHub stars corroborate the picture without setting the band.

Capability

4 high confidence
4.0

A full-coverage observability and evaluation platform: tracing, LLM-as-judge evaluation of both responses and retrieval, datasets and experiments, prompt management and a playground, cost and latency, and broad framework integrations. That covers every dimension this category grades and puts Phoenix level with the MLflow and Langfuse platforms that anchor 4.

  • https://github.com/Arize-ai/phoenix recorded 2026-08-13

    README feature list - Tracing "using OpenTelemetry-based instrumentation", Evaluation with "response and retrieval evals", Playground and Prompt Management, plus the arize-phoenix-evals and arize-phoenix-otel packages.

Verified 2026-08-13