YC Releases QM: An Open-Source Multiplayer AI Agent Harness for Collaborative Work
Y Combinator has open-sourced "qm", a multiplayer AI agent harness designed for collaborative work environments like Slack and the web. It introduces a unique contribution model where humans submit informal text proposals (ADRs) and AI agents implement the actual code. This project shifts the paradigm of open-source collaboration by having AI maintain the codebase while humans focus on high-level scoping and design. It addresses the complexity of scaling AI assistants from personal use to company-wide, collaborative environments. Built based on YC's internal experience running over 50 agents, QM utilizes per-person scopes and shared rooms to manage agent access. Notably, the project's contribution guidelines explicitly request human-written text proposals and discourage using AI to generate these proposals.
## BACKGROUND
An AI agent harness is the software scaffolding—including tools, memory, sandboxes, and guardrails—that wraps around a large language model to turn it into a functional agent. While traditional agents are designed as single-user personal assistants, multiplayer harnesses coordinate multiple users and agents working together.