QM provides a headless agent platform that runs both in Slack and on the web, built for startup-scale collaboration. Each user and each channel or project gets its own isolated sandbox—complete with memory, files, keychain, permissions, scheduled tasks, and custom web apps—so teammates can work independently or share skills without stepping on each other. Under the hood, the core is a TypeScript service on Node using Fastify for its HTTP API, backed by Postgres for sessions, memory, and queues. You swap in any of several model harnesses (Pi, OpenCode, Codex, Claude Code) without changing your deployment, and an execute tool runs commands in each scope’s durable container. Admins set organization-wide policies: choose which harnesses and models are allowed, define security postures (Strict, Auto, Dangerous), and control which skills and web apps people can access. Skills live in scopes and can be shared by grant; a promoted skill pack reaches the whole org. Background jobs (crons and watches) run tasks even when no one’s active. You get built-in search across notes, emails, documents, databases and the public web, inbox triage with drafts and labels, repository operations (tests, PRs, CI monitoring), and project tracking in channels with automated updates. To deploy, you run qm init with your org slug and target (Fly or AWS), then install dependencies. Your deployment directory holds all org-specific config, sandbox layers, tools, and plugin images; the core remains untouched upstream. You can also maintain a private fork that mirrors upstream with git clone and upstream remotes, separating core updates from your custom layers. Two helper skills—update-qm and upstream-pr—manage syncing in both directions, ensuring your private fork stays in step without exposing internal configurations. Contributions happen via text ADR files, and the security model and threat assumptions live in SECURITY.md.