yc-software/qm

What it solves

QM provides a multiplayer agent harness designed specifically for startups and organizations. Unlike personal AI assistants, it allows employees to have their own isolated workspaces while still enabling collaborative agent interactions in shared channels, group messages, and projects.

How it works

QM uses a headless core that interfaces with various LLM harnesses (such as Claude Code, OpenCode, or Pi) and a Postgres persistence layer for memory and session history. Each user or room has a scoped sandbox—a durable computer where tools are installed and executed. The system integrates with Slack and a web UI, ensuring a consistent identity and configuration across both platforms. It employs a strict security model with three postures (Strict, Auto, and Dangerous) to control tool execution and data screening.

Who it’s for

It is built for startups and teams that need a deployable, vendor-agnostic agent infrastructure that supports both individual productivity and organizational collaboration.

Highlights

• Multi-scope Memory: Separate, scoped memory, files, and permissions for individuals and shared rooms.
• ** webinars Web and Slack Integration:** Seamless identity and configuration carry-over between the web app and Slack.
• Customizable Skills: Support for shared skills that can be imported from git repositories or promoted org-wide by admins.
• Background Automation: Ability to run crons, watches, and inbound webhooks for autonomous work.
• Vendor Agnostic: Supports switching between different models and harnesses without being tied to a single provider.