Y Combinator has open-sourced QM (Quartermaster), an MIT-licensed multi-agent harness designed to facilitate collaborative work within organizations. Unlike traditional personal AI assistants, QM operates as a "multiplayer" system, running across Slack and the web to support diverse internal functions from accounting and legal to events and engineering. YC itself leverages QM internally, framing its public release as an early-stage experiment.
QM's architecture is built on the principle of isolated workspaces, providing each employee and collaborative room with its own scoped memory, files, permissions, and a durable sandbox. A central headless core manages APIs, identity, policy, and scheduling, driving an agent loop that interacts with a small, fixed tool surface. Crucially, one of these tools is an execute function, allowing commands to run within each scope's isolated sandbox. This design makes QM harness-agnostic, meaning it can be driven by various large language models like Pi, OpenCode, or Claude Code, without tying deployments to a single vendor.
This release is significant for developers and researchers seeking to implement sophisticated AI solutions in organizational settings. QM offers a robust framework for multi-agent collaboration, moving beyond the limitations of single-user assistants. It's deployable today for startups and mid-sized companies (10-500 people) comfortable with cloud infrastructure, allowing organizations to maintain full ownership and control within their own cloud accounts. Its flexibility in integrating different LLMs and providing secure, isolated environments makes it a powerful tool for building custom internal AI applications across various industries, from professional services to B2B SaaS internal tooling.