rekursiv.ai

rekursiv.ai builds autonomous research systems it calls AI Scientists. Instead of executing instructions the way a typical coding agent does, these systems form hypotheses, design and run experiments, review one another's results and report findings, with each claim traced back to its evidence. The company says its systems have already produced new machine learning methods and algorithms that outperformed the best published methods on several problems at a fraction of the usual cost. The product is described as a cockpit for autonomous research. A user defines a problem, and the platform dispatches a fleet of AI Scientist agents that set evaluations, run experiments, critique each other and push results to leaderboards. The interface shows live tasks, agent counts, best scores and reviewer commentary. It is aimed at research teams and organizations tackling machine learning and optimization problems that benefit from large scale automated experimentation. The company publishes a research blog covering results such as a reinforcement learning agent trained on the Craftax benchmark, a result on the NanoChat benchmark and work on a quantum circuit benchmark. Access is by request, through an email address on the site. No pricing or plans are published.