Phoenix helps you understand and improve AI applications by giving you a workflow for debugging and iteration. You can send detailed logging information, known as traces, from your app to see exactly what happened during a run, score outputs using evaluation tests to identify failures and regressions, iterate on your prompts using real production examples, and optimize your app with experiments that compare changes on the same inputs. Together, these tools help you move from inspecting individual runs to improving quality with evidence.
Phoenix is built by Arize AI and the open-source community. It is built on top of OpenTelemetry and is powered by OpenInference instrumentation. See Integrations for details.
In addition to Phoenix, Arize offers Arize AX, a managed enterprise platform built on the same open standards. Learn more about Arize AX and how the two platforms compare:
Features#
Tracing lets you see what happened during a single run of your AI application, step by step. A trace captures model calls, retrieval, tool use, and custom logic so you can debug behavior and understand where time is spent.
Phoenix accepts traces over OpenTelemetry (OTLP) and provides auto-instrumentation for popular frameworks (LlamaIndex, LangChain, DSPy, Mastra, Vercel AI SDK), providers (OpenAI, Bedrock, Anthropic), and languages (Python, TypeScript, Java).
Evaluations help you measure the output quality of your application. You can score traces & spans with LLM-based evaluators, code-based checks, or human labels so you can track performance and identify failures consistently. New to evaluations? Our AI agent handbook covers agent evaluation end to end.
- LLM-based evaluations — Run pre-built or custom evaluators on your data
- Dataset evaluators — Attach evaluators to datasets so they run automatically during experiments
- Evaluator integrations — Use Phoenix evals, or bring your own from Ragas, Deepeval, or Cleanlab
- Human annotations — Attach ground truth labels directly in the UI
Phoenix helps you iterate on prompts using real examples from your application. You can version prompts, test prompt variants across datasets, and replay calls to see how changes affect outputs before rolling them out.
- Prompt Management — Version, store, and deploy prompts
- Prompt Playground — Experiment with prompts and models side-by-side
- Span Replay — Debug by replaying LLM calls with different inputs
- Prompts in Code — Sync prompts across environments via SDK
Datasets & Experiments help you test changes systematically using the same inputs. You can group traces into datasets, rerun them through different versions of your application, and compare evaluation results to confirm whether a change actually improved performance.
- Run Experiments — Compare different versions of your application
- Create Datasets — Collect traces or upload from code/CSV
- Dataset Evaluators — Attach reusable evaluators to datasets as automated test cases
- Test at Scale — Run datasets through Playground or export for fine-tuning
Get started#
Start Phoenix, then work through Observe and Evaluate. Have an existing app? Let your coding agent set it up.
Other Resources#
Use PXI, the agent built into Phoenix, to debug traces and iterate on prompts in context
Add instrumentation for OpenAI, LangChain, LlamaIndex, and more
Example notebooks for tracing, evals, RAG analysis, and more
Join the Phoenix Slack to ask questions and connect with developers