Agent Performance Lab

Agent Performance Lab is a local-first Chrome DevTools extension for observing, testing, and governing browser-based AI agent workflows. It helps developers, QA engineers, and AI teams record opt-in browser sessions, review what happened, and turn useful agent runs into evidence, reports, baselines, and private eval suites. What you can do: - Record browser-visible agent workflows from the popup or DevTools panel. - Review timeline events, visible prompts and responses, network metadata, tool-like activity, possible MCP activity, governance findings, insights, and quality scores. - Flag risky prompt content such as password-like secrets and sensitive data. - Export structured JSON or Markdown reports for tickets, docs, reviews, and release notes. - Save successful sessions as local baseline test cases. - Generate local private eval suites from recorded sessions. Privacy: Agent Performance Lab stores recorded data locally in your browser. It does not upload sessions, prompts, responses, files, network metadata, governance findings, reports, or private eval suites to an external server. Recording is opt-in and starts only when you turn it on. Current beta limitations: Browser mode can only observe what Chrome and the inspected page expose. Backend model calls, server-side MCP calls, exact token usage, and hidden platform internals may be incomplete unless the agent platform exposes that data in the browser. Best for: - AI agent builders - QA and test automation teams - Developer tools teams - Security and governance reviewers - Teams creating repeatable private evals for browser-based agents