qm: Multiplayer AI Agents for Teams in Slack Channels

QM is a multiplayer agent harness designed for startups. Unlike traditional personal assistant agents, each employee has an isolated workspace while collaborating together in Slack channels or projects. It provides a structure that moves toward shared goals without affecting each other's work.

When a user passes a deployment address to their chosen coding agent, installation proceeds according to the repository's guidelines. You can freely switch between various harnesses and models, such as Pi, OpenCode, Codex, and Claude Code, avoiding vendor lock-in. All components are built on open-source foundations, ensuring high flexibility.

Each individual and room has a unique scope, with memory, files, keychains, permissions, cron jobs, web apps, and sandboxes managed independently. It performs actual development tasks, such as unified search across internal notes, emails, documents, databases, and the web, running tests in existing repositories, opening PRs, and monitoring CI. It also learns past writing styles to automatically generate inbox triage and reply drafts.

Security posture is selected from three options: Strict, Auto, and Dangerous, which can be tightened within narrower scopes. In Auto mode, a classifier filters external data and tool results before passing them to the model. All tool calls are audited, and dangerous commands such as recursive deletions or destructive SQL are blocked according to predefined policies in any posture.