Skip to main content

> ML_LIBRARY // ARIZE-PHOENIX_v1.0

Arize Phoenix

Arize AI — Open-source AI observability, LLM tracing, and RAG evaluation platform built on OpenTelemetry.

evaluation-observabilityv4.15.0ELv2 (Permissive for internal enterprise use)qualified
Official DocumentationSource Repository

Model Training

Not Supported

This library is a dedicated runtime engine for inference serving and does not train models.

Model Inference

Supported
Inference Accelerators:
CPU
Deployment Targets:server

What It Does

  • +Distributed tracing for LLM and agentic workflows compliant with OpenTelemetry and OpenInference standards
  • +Out-of-the-box evaluations for RAG: Hallucination detection, Question-Answer correctness, and Retrieval precision
  • +Interactive 3D UMAP embedding visualizer to diagnose retrieval blindspots and topic drift
  • +Local in-notebook UI launching with a single python call (px.launch()) or deployable via Docker/Helm

What It Does Not Do

  • -Train foundation neural weights or fine-tune LLMs directly
  • -Replace core infrastructure APM metrics (Datadog/Prometheus)
  • -Be resold as a competing managed SaaS service (ELv2 restriction)

>Suitable Work Types

  • Debugging multi-agent LangChain, LlamaIndex, or DSPy tool-call traces
  • Evaluating retrieval chunk quality and hallucination rates across enterprise RAG pipelines
  • Visualizing vector embedding clusters to discover user query blindspots

>Unsuitable Work Types

  • Classical tabular gradient boosting training
  • Low-level CUDA kernel profiling
Data Residency Implications

Runs 100% locally or inside a self-hosted private Kubernetes cluster. Zero telemetry sent to Arize cloud.

Security Considerations

ELv2 license allows free internal enterprise use and embedding into SaaS products as long as Phoenix is not offered as a standalone managed service.

Operational Profile & Known Limitations

Maturity:mature
Learning Curve:low
Ops Complexity:low
Cost Tier:free-oss
> Known Limitations:
  • ELv2 license prevents reselling Phoenix as a commercial cloud evaluation service to third parties.

Associated Incident Patterns (Incidentpedia)

Enforce safeguards and monitoring to guard against these documented real-world failure modes:

thread-pool-exhaustion

> Primary Evidence & Benchmark Citations

Arize Phoenix Documentationofficial-docs • >=4.10.0, <=4.15.x
2026-09-25HIGH