May 08, 2026
Arize Phoenix is the open-source answer to the need for deep visibility into AI systems. It provides a notebook-first experience for tracing LLM calls and evaluating the performance of your RAG pipelines.
Phoenix allows you to visualize the full execution trace of your application, from the initial prompt to the final response, including every tool call and retrieval step. You can see the exact "latency" of each step and identify bottlenecks in your agent's reasoning process, making it much easier to optimize for speed and reliability.
One of Phoenix's unique features is its ability to visualize your embeddings in a 3D space. By mapping your data and your model's failures, you can identify "clusters" where the model is struggling. This "visual debugging" allows you to see if your RAG system is failing because of poor retrieval or because the model doesn't understand a specific subset of your knowledge base.