Observing systems around a model
Arize AI builds tools for inspecting AI applications and evaluating their outputs. Its Phoenix project supports tracing, evaluation and local deployment, while Arize AX is the managed platform. Application traces and scores provide evidence about recorded behavior; they do not by themselves reveal a model’s internal reasoning or establish that every answer is true. [1]
The Observe introduction
Arize introduced Phoenix at Observe in April 2023, describing notebook-based embedding analysis and clusters of problematic outputs. The event records that introduction and reported availability, rather than claiming it was the first public code release. [2]
Read the actual license
Phoenix’s current license document specifies Elastic License 2.0. The availability of source and self-hosting is useful to teams choosing where application data is processed, but a source-available license and an unrestricted right to offer a competing hosted service are different permissions. [3]
Sources
- Arize Phoenix: Open-Source LLM Observability & Agent Evaluation
Arize AI · Oct 7, 2026
- Arize Debuts Phoenix, the First Open Source Library for Evaluating Large Language Models
Arize AI · Apr 25, 2023
- Phoenix Elastic License 2.0
Arize AI · retrieved Oct 23, 2026