Debugging and evaluating AI systems requires visibility into every step. Arize Phoenix is an open-source observability and evaluation platform, with OpenTelemetry-based tracing that works with LangChain, LlamaIndex, OpenAI and more, all self-hostable and vendor-neutral.
What is Phoenix?
Phoenix does OpenInference-standard tracing (over OpenTelemetry), LLM-as-judge evaluations, dataset and experiment tracking and a prompt playground. It runs locally or in a container, which helps in privacy-sensitive workflows.
Key features
- OpenInference/OpenTelemetry tracing, framework-agnostic
- LLM-as-judge evaluations and experiments
- Dataset tracking and prompt playground
- Self-hostable (notebook, container or app)
How Reche uses it
Observing and evaluating is what turns an AI system from a gamble into engineering. Reche instruments products with tools like Phoenix, so quality and cost are measured and improved, not estimated in the dark.