Research project

Autonomous Repository Investigator

An AI agent that investigates a repository's commits, pull requests, issues, and CI runs to answer open-ended questions about it — and cites its evidence the way an engineer would.

Across six models and three real repositories, giving the agent a small set of curated tools beat giving it open access to write and run its own code, for every model tested.

Try it live See the findings

Ask

Pick a repo and question, watch the agent investigate it live, step by step.

Compare

See the curated-tools and open-sandbox agents answer the same question, side by side.

Findings

The study's headline results: which approach produced better answers, and at what cost.