---
type: "post"
stable_id: "post:qm-multiplayer-agent-harness"
slug: "qm-multiplayer-agent-harness"
title: "QM Turns Team Agents Into Scoped Runtime State"
description: "QM points at a multiplayer agent harness where identity, memory, permissions, credentials, crons, and sandboxes are scoped by person, channel, and project."
retrieval_nugget: "QM points at a multiplayer agent harness where identity, memory, permissions, credentials, crons, and sandboxes are scoped by person, channel, and project. QM is interesting because it frames the company agent as a runtime, not a bot in a group chat. The repository describes a multiplayer agent harness for Slack and web use, with scoped memory, files, keychain access, permissions, crons, web apps, and durable sandbox"
status: "published"
published_at: "2026-08-04"
updated_at: "2026-08-04"
record_date: "2026-08-04"
date_kind: "published_at"
topics: ["coding-agents","operations","developer-tools","slack"]
source_urls: ["https://github.com/yc-software/qm"]
visuals: [{"id":"qm-multiplayer-agent-harness","kind":"editorial-diagram","role":"hero","src":"/images/drip/qm-multiplayer-agent-harness/qm-multiplayer-agent-harness.webp","alt":"Hand-drawn hub-and-spoke runtime map with a headless core, storage, scoped workspaces, memory, keychains, permission gates, crons, sandboxes, and adapters.","caption":"New Runtime synthesis: multiplayer agents need scoped runtime state, not just a shared transcript.","credit":"New Runtime synthesis from public source material","source_url":"https://github.com/yc-software/qm","generated_with":"gemini-3.1-flash-image","width":1600,"height":900,"legend":[{"label":"Scope","description":"Each person, channel, or project needs its own memory and permission context."},{"label":"Core","description":"The harness coordinates state, credentials, queues, and durable execution."},{"label":"Sandbox","description":"Work runs in recoverable environments instead of disappearing into chat history."}]}]
schema_version: "newruntime-agent-readable-v0.2"
routes: {"html":"https://newruntime.com/posts/qm-multiplayer-agent-harness/","markdown":"https://newruntime.com/posts/qm-multiplayer-agent-harness.md","json":"https://newruntime.com/posts/qm-multiplayer-agent-harness.json"}
---

# QM Turns Team Agents Into Scoped Runtime State

## Retrieval answer

QM points at a multiplayer agent harness where identity, memory, permissions, credentials, crons, and sandboxes are scoped by person, channel, and project. QM is interesting because it frames the company agent as a runtime, not a bot in a group chat. The repository describes a multiplayer agent harness for Slack and web use, with scoped memory, files, keychain access, permissions, crons, web apps, and durable sandbox

QM is interesting because it frames the company agent as a runtime, not a bot in a group chat. The repository describes a multiplayer agent harness for Slack and web use, with scoped memory, files, keychain access, permissions, crons, web apps, and durable sandboxes.

That is the right shape for organizational work. A team agent cannot treat every user, channel, project, credential, and long-running task as one flat transcript. The operating unit has to be scoped.

This is close to the OpenClaw and Hermes boundary. One system can stay heavy and approval-gated; another can run the fast operational lane. What matters is that identity, memory, permissions, and receipts are explicit before the agent starts acting.

For New Runtime, QM is a useful reference point for the next layer: agent work should become multiplayer only after the runtime can say who asked, what scope was active, which tools were allowed, and where the durable state lives.