QM Releases AI Agent for Teams

QM is an open-source multiplayer AI agent harness that multiple members use together on Slack and the web. It separates workspaces by person and chat room, independently managing memory, files, keychain views, permissions, cron, web apps, and sandboxes.

It solves the configuration conflicts, shared credentials, and mixed conversation memory that arise from shared bot accounts through scope-based isolation. Each scope's sandbox operates not as a disposable container but as a persistent execution environment where tools and state are maintained.

• Supported harnesses/models: Pi, OpenCode, Codex, Claude Code
• Touchpoints: Slack channel/group messages and web UI
• Execution: the execute tool in a scope-dedicated sandbox
• Architecture: Node/TypeScript core, Fastify API, Slack Bolt plugin, Vite/Lit web UI, Postgres persistence layer
• Deployment: the qm CLI validates and deploys a deployment directory containing org settings, custom tools, skills, sandbox images, and infrastructure

The core handles the API, identity, policy, scheduler, and agent loop, and separates underlying elements into interfaces so that harnesses, session storage, sandbox, and memory implementations can be swapped out. The public repository is yc-software/qm, and the version at the time of writing is v0.1.4.

It primarily targets teams that use Slack and can deploy directly to their own Fly or AWS account, especially startups that want to directly manage in-house data and security settings. One consideration for adoption is that it's still an early-stage project at the v0.1.x stage.