Y Combinator open-sourced QM on July 31, a multi-agent harness the accelerator built and now runs internally across accounting, legal, events and engineering — including the work of building QM itself. The project is released under an MIT license, with YC describing it as easy to customize like Hermes or OpenClaw, but designed for company-wide use rather than a single individual.

What QM does

QM is pitched as a "multiplayer agent harness for work": every employee gets their own scoped workspace — memory, files, a keychain view, permissions, crons and a durable sandbox — while also being able to collaborate with agents inside shared Slack channels, group messages and projects. The system is cloud-first with native Slack and web UI, and it doesn't lock organizations into one model or coding harness — teams can switch between Pi, OpenCode, Codex and Claude Code depending on the task. Administrators get org-level controls, including a configurable security posture (Strict, Auto or Dangerous mode) with content screening and approval workflows, and the underlying stack runs TypeScript on Node with Fastify and Postgres.

Reception

The release landed quickly with developers: the GitHub repository passed 5,000 stars within its first day, and the project reached the #2 spot on Hacker News with over 500 points. Much of the discussion has focused on QM's per-person and per-room scoping model, which YC frames as the key difference from single-user coding-agent tools — it's built for teams working alongside agents in shared channels rather than one developer driving one assistant.

Why it matters

QM's release adds to a fast-growing category of "agent operating systems" that sit on top of existing model harnesses rather than compete with them directly, letting organizations standardize workflows, permissions and audit trails while remaining free to swap the underlying coding or chat agent. Coming from YC — which has direct visibility into how thousands of startups are actually deploying agents day to day — the tool's design choices, and its rapid uptake among developers, are likely to influence how other companies think about deploying agents beyond single-user assistants.