What problem does it solve?
Phoenix helps you debug and improve LLM applications by turning opaque model calls into searchable traces, repeatable evaluations, and real-time monitoring signals.
Core Features & Use Cases
- End-to-end LLM tracing with OpenTelemetry: capture traces/spans across frameworks (OpenAI, LangChain, LlamaIndex, Anthropic) so you can pinpoint where quality or latency issues originate.
- Evaluation framework (LLM-as-judge): run automated quality checks (e.g., hallucination, relevance, toxicity) against datasets and log results.
- Datasets, experiments, and feedback: version datasets for regression testing, run A/B or model-comparison experiments, and attach user feedback/annotations for investigation.
- Self-hosted observability: operate with PostgreSQL or SQLite while avoiding vendor lock-in.
Use Phoenix when you need to systematically measure prompt/model changes, investigate production issues with trace-level context, or continuously evaluate LLM outputs over time—then compare versions through experiments and dashboards.
Quick Start
Install Phoenix, launch the server, and instrument your LLM client so traces begin appearing in the Phoenix UI for project_name "my-llm-app".
