Drums is building infrastructure for computer-use RL, so AI labs can run more training attempts on less hardware.

We're building computer gyms around a shared application runtime for computer-use training. Instead of giving every training attempt a full computer of its own, Drums lets many attempts share the same running OS and applications while keeping each attempt's mutable state separate, including files, browser sessions, and application changes. Each attempt must run and reset without affecting any other attempt's state. The goal is simple: cut duplicated resource use, run more training on the same hardware, and lower the cost of training environments.

Run your RL here →Invited teams onlyRequest accessDocsResearchBlog