QM: A Multiplayer Agent Harness for Teams

QM is an open-source multiplayer agent harness designed for team environments. Unlike personal assistant agents that can become complex when scaled across groups, QM provides isolated workspaces for individuals while supporting collaboration in channels, group messages, and projects.

Core Architecture and Capabilities

• Independent Workspaces: Every user and room has a scoped memory, file storage, keychain view, permissions, crons, web apps, and a durable sandbox.
• Always-On Operation: Agents run independently on schedules and webhooks in your chosen cloud environment, using models and API keys of your choice rather than being tied to a single vendor.
• Dual Interface: Supports interaction via Slack and a web application sharing the same identity and configuration.
• Customization and Apps: Administrators can set organization-level security configurations, model availability, and publish custom internal web apps.

Technical Implementation

QM core is built with TypeScript on Node.js using Fastify for HTTP routing. It relies on a PostgreSQL persistence layer for user data, session history, and state management. The architecture includes an execute tool that runs commands inside an isolated sandbox environment.

Security and Deployment

Operators deploy QM in their own cloud accounts using providers like DigitalOcean, E2B, or Render. The project outlines multiple security postures:

• Strict: Requires human approval for every tool call, excluding turn enders.
• Auto (Default): Blocks private-network access.
• Dangerous: Permits tool execution without approval gates.

Content screening and command policies can be further configured by administrators to restrict potentially destructive operations.

Verification and Resources

Documentation, deployment guides, and source code are available via the official GitHub repository and documentation site at labs.ycombinator.com/qm. QM is available under the MIT License.