Y Combinator Open-Sources QM: AI Agent Harness for Work

Y Combinator has released QM, an AI agent harness it uses to run itself. This tool is now open-source, available on GitHub under the MIT license. What's interesting is that you can inspect the system directly instead of relying on a demo. You can fork it, modify it, and even run it commercially. QM is designed as a multiplayer agent harness for work, built for both Slack and the web. It provides each user and each room with its own isolated memory, files, permissions, and a durable sandbox. This design tackles the challenge of context and credentials when multiple people use the same agent system. Y Combinator developed QM after experiencing difficulties managing a fleet of over 50 internal agents. The name QM, short for quartermaster, reflects its role in coordinating the operational layer of agents. The core of QM runs on TypeScript with Node, uses Fastify for HTTP, and stores session history in Postgres. It also offers Slack as an optional plugin. Crucially, QM is not tied to a single AI model or coding tool. It can be driven by various models like Pi, OpenCode, Codex, and Claude Code. This flexibility is important for companies building workflows today.