Y Combinator open-sources QM, the multiplayer agent harness it runs its own back office on

Y Combinator open-sourced QM, a multiplayer agent harness it built to run its own back office, on July 31, 2026, releasing it under an MIT license on GitHub.

The startup accelerator says it uses QM internally to run parts of its own operations. Rather than a single chatbot, the system gives each employee and each Slack room its own scoped memory, permissions, keychain view, scheduled jobs and durable sandbox, so one agent identity follows a user between Slack and a web interface.

QM is designed to run other coding agents rather than replace them. Its documentation says it works with multiple agent loops — including Pi, OpenCode, Codex and Claude Code — and offers Strict, Auto and Dangerous security postures that govern how much an agent can do without a human in the loop. The stack is TypeScript on Node with Fastify, Postgres for session state, Slack’s Bolt SDK and a Vite and Lit web UI.

The release landed hard among developers. The GitHub repository quickly drew more than 4,000 stars, becoming one of the most-discussed new open-source projects that day.

Two caveats temper the enthusiasm. YC’s account of running its own operations on QM is the company’s own; there is no third-party audit of how much the accelerator actually delegates to agents, or how the Dangerous posture behaves in production. An MIT license lowers the bar to adoption, but a harness built for one firm’s workflows is not guaranteed to fit another’s.