Y Combinator Open-Sources QM, the AI Agent Harness It Uses to Run Itself

Aug 1, 2026

Y Combinator has open-sourced QM, the multiplayer AI agent harness it built and now runs internally across its own accounting, legal, events, and engineering teams. Released under an MIT license on GitHub under the yc-software organization, QM gives every employee and every Slack room its own scoped memory, permissions, and sandbox, and it works with Pi, OpenCode, Codex, or Claude Code.

QM is described as a multiplayer agent harness for work, built for Slack and the web, with scoped memory, files, keychain view, permissions, crons, web apps, and durable sandbox per person and room.

Background

YC previously experimented with a basic Ruby agent loop, then provisioned more than 50 Hermes agents for individual employees. Managing that fleet became difficult; OpenClaw pushed the team toward QM (short for quartermaster).

Core stack (per README): TypeScript on Node, Fastify for HTTP, Postgres for session history and memory, Slack as optional in-process plugin.

Model-agnostic design: Pi, OpenCode, Codex, and Claude Code can all drive the same core. Security model follows local coding agents—agent acts as the person it works for, with audited actions; postures range from strict human approval to a “dangerous mode” with no pauses between tool calls.

Repository: https://github.com/yc-software/qm