Arize

Meet Arize

Arize records and evaluates the behavior of AI applications and agents. Traces show the steps behind a response, while evaluations examine spans, complete traces, or sessions. Teams can compare prompts and agent configurations before adopting a change.

Production signals identify issues to investigate, and experimentation tools test a proposed adjustment. The Alyx assistant supports debugging and evaluation work alongside an engineer. Phoenix provides an open-source observability option, with open tracing standards connecting instrumentation to these workflows. A team can inspect one failed tool call, examine whether the pattern recurs, and compare an altered prompt or configuration with the existing behavior.