Layerbrain

Super Mario Kart is now in WarGames. Same harness, new game. Agents drive from pixels with keyboard only. Race missions across every shipped track, from simple laps to harder races with more karts. github: github.com/layerbrain/war…

Today we’re releasing WarGames: a real-time RL environment where agents play Command & Conquer: Red Alert from pixels only. The agent sees the screen and acts with mouse and keyboard. Episodes run for hundreds or thousands of actions. Rewards come from hidden simulator state. Built for long-horizon vision-first agents that need to plan and adapt while the world changes. repo: github.com/layerbrain/war… post: layerbrain.com/index/war-games

There is no Trunks server. The whole product is a CLI and a tiny git shim. When you git push, the shim catches the call and writes the bytes straight to your storage. Another machine runs trunks pull and reads from the same place. No daemon. No control plane. No hosted middleman. Your bucket is the backbone. Github: github.com/layerbrain/tru…

Git push, to anywhere. Trunks turns any backend into a git remote. S3, R2, Tigris, MinIO, Postgres, SFTP, NFS, SMB, local disk, or a file share. Use git like normal. Trunks does the rest. Built for sandboxes, parallel agent runs, and handoffs before work becomes a GitHub PR.

Flight simulator is now in WarGames. Same harness, new game. Agents fly a Cessna 172 from pixels with mouse and keyboard only. Tutorial missions from preflight to landing. github.com/layerbrain/war…

Also works with @PrimeIntellect. Use WarGames for evals or RL training with Prime RL, or bring your own trainer. WarGames handles episodes, pixels, actions, and rewards. Prime env: app.primeintellect.ai/dashboard/envi…

introducing world operating systems. a theoretical framework for generative software. a learned runtime that watches the machine and outputs two streams at once: what you see on screen and what the computer actually does. read more here: layerbrain.com/index/world-op…