Applied AI systems
We focus on turning useful AI workflows into products that can operate reliably in real environments, not only in demos.
Research
Research
We are interested in the systems research behind agents that are easy to use at their full potential: production-ready agents, efficient infrastructure, provider-independent execution, and reliable multi-step task completion.
Efficient compute
Alysis explores how to run inference and training workloads efficiently, keeping agents fast and responsive without locking you to a single provider or model.
Agent reliability
Our work examines how to make AI agents plan, run work in parallel on isolated git worktrees, and verify results before merging — the research behind Forge Mode — so they hold up on real-world work.
Provider-independent agents
We study how to make powerful agents easy for anyone to use, and how to keep them from being locked to the single provider or model that built them — so the best agent works everywhere.