- Published on
An MIT-Licensed Multiplayer Agent Harness That Runs In Slack And The Web
- Authors
- Name
- DeepHorus
Y Combinator has open-sourced QM, the multi-agent harness it uses internally. The software, also called quartermaster, is designed for work collaboration and runs in Slack and on the web. YC has been running QM across accounting, legal, events, and engineering, including building QM itself. The entire project ships under the MIT license.
QM is not a typical personal assistant. The YC team argues that stretching one assistant across a whole company quickly becomes complex. Instead, QM gives each employee an isolated workspace that others cannot affect. People can also collaborate with agents in channels, group messages, and projects.
Each person and room gets its own scoped memory, files, keychain, permissions, crons, web apps, and durable sandbox. The same identity carries between Slack and the web app. Skills are shareable by grant and can be promoted to the whole org. This architecture supports applications like searching internal notes, email, documents, databases, and the web together, triaging inboxes, running tests, opening PRs, and tracking projects.
QM is deployable today, but it requires cloud infrastructure and a platform engineer. The best fit is a startup or mid-sized company of 10 to 500 people. Larger enterprises can run it but should review security first. Realistic industries include venture capital, professional services, fintech, legal operations, events, and B2B SaaS.
Technically, every turn runs through a central headless core that handles API, identity, policy, and scheduling. A Postgres layer holds durable state, and the agent has a small fixed tool surface including an execute tool for isolated sandboxes. QM is harness-agnostic, supporting Pi, OpenCode, Codex, and Claude Code, so deployments are not tied to one vendor.
The security model follows local coding agents, where the agent acts as the person it works for, with audited credentials and permissions. Orgs can choose strict pauses, a default auto mode with classifier screening, or a dangerous mode. YC frames the release as an early experiment with bugs, but it is open for anyone to deploy.
Looking ahead, QM’s open-source nature could accelerate adoption of multi-agent workflows in companies. By decoupling from a specific model vendor, it offers flexibility. The YC team expects the community to contribute skills and improvements. The project is available now under MIT license.
