Loopr

Loopr — A recursive AI testing agent that finds bugs, fixes them, and loops until your code is clean. Writing tests is slow. Fixing bugs is slower. Doing both repeatedly until everything passes? Nobody has time for that. Existing tools generate tests once and stop — they don't fix failures, don't learn your codebase, and don't remember what they already tried. Loopr does. Point it at a Python repo and it enters a recursive loop: generates pytest functions with IBM watsonx.ai Granite, runs them, suggests targeted fixes for failures, shows a colored diff for your approval, applies the fix, and repeats until all tests pass. Every bug found and every fix applied is stored in a local context bank (.agent-context.json) — so Loopr never repeats the same mistake and gets smarter about your codebase across runs. Run 1 finds the obvious bugs. Run 2 generates smarter tests informed by what it already learned. Run 3 exits clean. Built with IBM Bob, powered by IBM watsonx.ai. Bob acted as the development partner across every module — meaning Loopr was itself built by an agentic IBM workflow, demonstrating its own premise. Granite handles runtime reasoning, test generation, and fix suggestions. Install with pip install -e ., run with loopr ./your-repo, and walk away.