Y Combinator open-sourced QM on July 31, 2026 — an MIT-licensed multi-agent harness the accelerator built and used internally across its accounting, legal, events, and engineering teams, now available for any company to self-host at no cost.
QM gives every employee their own isolated AI workspace with scoped memory, files, keychains, permissions, cron jobs, web apps, and a sandbox — preventing one user’s agent context from leaking into a colleague’s. The system integrates natively with Slack and a web UI. YC published the code on GitHub at yc-software/qm under the MIT license. QM runs on Postgres for session, memory, and queue persistence. The harness is model-agnostic and supports Pi, OpenCode, OpenAI Codex, and Claude Code driving the same core infrastructure.
What QM Does That Most AI Tooling Does Not
QM solves the organizational isolation problem that individual AI assistants such as ChatGPT and Claude cannot: shared work happens in team Slack channels while each person retains a personalized, permission-bounded workspace. Admins set org-level configuration, define which harnesses and models are available, and choose a security posture for the entire organization.
QM ships with three security postures. Strict pauses every tool call for human approval before execution. Auto — the default — uses a classifier to screen all external data before agents act on it. Dangerous disables all screening. Agents run under the employee’s own credentials, creating an audit trail tied to the individual rather than a shared service account.
The closest analogues the README identifies are Hermes and OpenClaw, both developer-focused coding harnesses not designed for general business use. QM’s differentiators are per-user isolated sandboxes, native Slack integration, admin-level security posture controls, and multi-vendor model support. Commercial alternatives such as HubSpot Agent Hub and OpenAI Presence are SaaS products tied to single vendors; QM is self-hosted and vendor-independent.
Why YC Built It and Why the Source Release Matters
YC is a small organization that operates at outsized output — a point YC staff member Eve Bouff made explicit at launch: “i’ve been the #1 power user of qm, our internal multi-agent harness at yc, for months. today, we’re open-sourcing it. it’s been mesmerizing watching yc partners and staff use it to 1000x themselves.” YC’s official account described QM as “meant to be easy to customize, like Hermes or OpenClaw, but useful for a whole company.”
Before QM, teams adopting AI org-wide faced a credential-sharing problem: single-user setups could not scale without giving multiple employees access to the same API key or agent account. QM addresses this with per-employee sandboxes while preserving shared-channel collaboration. For businesses evaluating best AI agents for business, QM is the first open-source option designed explicitly for company-wide deployment rather than individual developer use.
How to Self-Host QM
QM is cloud-first. Deployment requires a Postgres instance for persistence and configuration of the desired model integrations (Claude Code, Codex, or Pi). Admin controls set which harnesses and models employees can access. Source code, setup instructions, and security documentation are published at github.com/yc-software/qm. Compute and infrastructure costs apply; QM itself is free under the MIT license, but running agents at scale requires cloud or on-premises compute.
Our Take
QM is the most credible open-source answer yet to the “AI for my whole company” question, for one reason: YC already proved it works at an organization known for doing more with fewer people than almost anyone. The real competition is not other agent harnesses — it is the SaaS vendors who charge per seat for something a startup can now self-host for the cost of infrastructure. Any sub-200-person company that has been waiting for an enterprise-grade, model-agnostic harness without a vendor lock-in should evaluate QM before the paid alternatives.
For Context
Team-based AI agent deployment has been a growing category in 2026. Buzz by Jack Dorsey introduced AI agents inside team chat channels earlier this year. The MCP 2026 spec standardized the protocol that many agent harnesses — including the tools QM supports — use to connect models to external tools and data sources.
Related
- 12 Best AI Agents for Business Tasks — updated guide to the top options
- HubSpot Agent Hub — CRM teams running agents in shared workspaces
- Buzz by Jack Dorsey — agents embedded in team chat
