Arize is an AI engineering platform for observing, evaluating and improving production AI agents. Its product, Arize AX, traces what an agent did, measures whether it is getting better or worse, and helps developers test prompt and harness changes before they ship.

Key features:

Trace agents with OpenInference, an OpenTelemetry-based standard, including multimodal traces.
Run span, trace and session-level evals, online or on datasets.
Debug and fix agents with Alyx, an AI engineering agent.
Work with traces from coding agents such as Cursor, Claude Code and OpenCode.
Replay traces in a playground, compare prompts and optimize them.
Store traces in ADB and connect them to BigQuery, Databricks or Snowflake.

Arize lists SOC 2, ISO, GDPR and HIPAA compliance, with data regions in the US, EU or Canada, and offers Phoenix as a free open-source tool. AX Free includes 25k spans a month with unlimited users and evals, AX Pro is $50/month, and Enterprise adds self-hosting and SSO.