Move 37 is a reference to AlphaGo's unconventional move in its match against Lee Sedol: a play that first looked strange, then proved decisive. We use the name because we want to help create more moments like that for complex, critical problems.
We are an independent AI research lab focused on building self-improving systems. Right now we are exploring two connected problems.
The first is AI research itself: systems that can read current work, set up experiments, improve training loops, and make it easier to turn feedback into better models.
The second is physical AI and reinforcement learning for physical systems. We want AI that can reason about the real world, learn from interaction, and improve in environments where feedback is noisy and expensive.
In the end, we are trying to build AI that builds better AI, starting the snowball effect. We build and open source tools that help us get there.
Tools, systems, and research projects from the lab.
- Paperbase: CLI for AI research papers
- Edgent: Headless browser agent
- Rollout: Feedback loops that make agents better over time
- Workdir: Low-cost Firecracker sandboxes for agent workloads