QM is an open-source multiplayer agent harness designed to help startups deploy AI agents across their teams. According to the GitHub repository, most agents are built as personal assistants, which becomes complex when scaled to entire organizations. QM addresses this by giving each employee their own isolated workspace while enabling collaborative work through Slack channels, group messages, and projects.
The system provides personal and shared scopes, allowing employees to customize their agent interactions individually while still collaborating on shared tasks. Each person and room has its own scoped memory, files, keychain view, permissions, scheduled tasks (crons), web apps, and durable sandbox environment.
A key design principle is vendor independence. QM’s core supports multiple models and harnesses including Pi, OpenCode, Codex, and Claude Code, so deployments are not locked into any single vendor. Users can switch between harnesses without rewriting their setup.
The platform includes several built-in capabilities: search across internal notes, email, documents, databases, and the web; retrieval from company knowledge bases; creation and publication of custom internal apps; email triage based on writing voice patterns; work within existing repositories including running tests and monitoring CI; and background task execution through scheduled crons and watches.
Security is configurable through three postures. Strict mode pauses every tool call for human approval. Auto mode (default) screens external data before it reaches the model. Dangerous mode removes content screening entirely. Regardless of posture, predeclared command policies apply hard denials for destructive actions like recursive deletes.
The architecture uses a Postgres persistence layer for sessions and memory, a Node.js core running TypeScript, and optional plugins for Slack (using Bolt), web UI (built with Vite and Lit), and admin panels. Each agent scope gets its own isolated sandbox with durable state—installed tools remain available across sessions.
QM follows a local coding agent model where the agent acts with the person’s credentials and permissions, with all actions audited. Organizations can deploy QM to their own cloud accounts (Fly or AWS) and keep customizations private while tracking upstream improvements. The project accepts contributions as human-written proposals rather than code submissions.
Key facts
- QM is designed for startups and supports multiple AI models including Claude Code, OpenCode, Codex, and Pi
- Each employee gets an isolated workspace with scoped memory, files, permissions, and sandboxes while enabling team collaboration
- The system integrates with Slack and web interfaces, with configuration carrying between both platforms
- Three security postures are available: Strict (approval-gated), Auto (content screening), and Dangerous (no screening)
- QM is open-source under MIT License and available on GitHub