YC Software released QM, a self-hosted platform for running AI agents across an organization rather than as one person's personal assistant, according to the project's GitHub repository.
"Each person and each room has its own scoped memory, files, keychain view, permissions, crons, web apps, and durable sandbox," the repository states. Employees and shared rooms both get isolated workspaces, and organizations can build custom internal web apps on top of an agent's output and control who can see them.
QM runs on TypeScript and Node with a Fastify server and Postgres for persistence, and plugs into multiple agent harnesses, including Pi, OpenCode, Codex and Claude Code, rather than locking a company into one model provider, according to the repository. Admins can set security postures ranging from requiring manual approval for every action to automatic screening.
Organizations deploy QM into their own cloud accounts through a command-line tool, without needing to check out the source themselves, the repository states. The project has drawn 14,500 stars and is licensed under MIT.
Most agent tools on the market are still built around a single user talking to a single assistant. QM's bet is that the harder, more valuable problem is giving a whole team of employees and agents shared, permissioned workspaces without collapsing everyone's memory and files into one undifferentiated pile.