← Portfolio · Portfolio case · Developer tools, coding agents · Switzerland
LogicStar: self-healing software
Seed investor · logicstar.ai · All companies
LogicStar builds an AI agent for the maintenance of commercial applications: it finds, reproduces, fixes and tests bugs autonomously, each fix arriving as a pull request with a regression test.
Why we invested
An ETH Zurich spin-out with the research depth to push the state of the art in automated code repair and the discipline to measure it publicly. The moat is expertise: a small team of world experts building a critical part of the new AI stack, verifiable on public benchmarks.
Milestones
- September 2025. New record on SWE-Bench Verified: 76.8% accuracy and 80% precision in automated code repair (post).
- May 2026. CodeTaste, a large-scale benchmark for AI coding agents on real refactoring tasks built with ETH Zurich, accepted to ICML 2026.
- Summer 2026. Published 38 findings on a public EU AI Act RAG codebase: 37 closed, 19 fixed and merged the same day through a single pull request with regression tests, every issue verifiable because the repository is public. Startup Partner at the AI+X Summit 2026.
Milestones are compiled from the company's own announcements and our Dispatch newsletter; figures are the company's.