YC Just Open-Sourced Its Multiplayer AI Agent
Por Miguel Torrez, fundador de Genial Agency
You will have QM, YC's open-source multiplayer AI agent, running for your team with an API key, and you will know where each workspace and admin feature lives (keychain, apps, judgments, governance, audit).
Lo que necesitas
A coding agent such as Claude Code or Codex to run the install
An API key for a model (QM does not accept your ChatGPT or Claude subscription yet)
Optional: a Slack workspace, if you want QM to live in Slack too
Paso a paso
Understand what QM is
QM is YC's open-source agent harness, similar to Hermes Agent or OpenClaw, but built for teams. It lives in a web app and in Slack at the same time, and each project (personal or shared) gets its own memory, files and skills. Cron jobs, skills and memory are included.
QM GitHub repohttps://github.com/yc-software/qm
QM websitehttps://qm.ycombinator.com
Install it with your coding agent
Give the repo link to Claude Code or Codex and ask it to run QM for you. Expect a few changes to run it locally: QM asks for Slack access right away, so tell your agent to skip the Slack connection if you will not use it.
Prompt for your coding agentHey, please run this QM instance for me.
Plug in a model through an API key
QM is API-based only: you cannot use your subscription out of the box. Add an API key. I recommend open-source models such as DeepSeek V4 to keep costs down. If you want to use your OpenAI subscription, ask your coding agent to modify QM for it.
Create a project and test the agent
Open the web app. Create a personal project, or a project shared with teammates. Start a new chat and send a test message to check the model responds.
Test messageHey, QM, how are you doing?
Set up files, crons, keychain and apps
In your user view: choose whether each file is shared only with the agent or with teammates too. Add cron jobs for scheduled tasks. Use Keychain to share credentials without exposing the key itself. Open Applications to find apps teammates built and use the deploy button at the top.
Review activity in the admin dashboard
Open the admin dashboard to see every session, from the web app and Slack. Open the Judgments tab to see what agents did and the reasons they gave. Check the preloaded skills (including the onboarding skill), organization memory, deployments and everyone's crons.
Track usage in Metrics
Open Metrics to see how many people use the agent, total tokens used and response times.
Set governance and data boundaries
Use Onboarding to add new users. In Governance, set model permissions, credentials and data boundaries so each person only reaches the data they should. Share connectors, manage users and the keychain, then open Audit to see every event and tool call run by members.
Add what is missing
QM is open source, so add any missing feature yourself: ask QM itself, Codex or Claude Code to make the change. My first addition would be subscription login so the agent can run on a subscription instead of API keys.
Ten cuidado con
QM runs on API keys only: your ChatGPT or Claude subscription will not work unless you (or your coding agent) modify it.
Running locally needs a few changes because QM asks for Slack access immediately; skip it if you are not using Slack.
Do not share API keys in chats: use the Keychain so teammates can use a credential without seeing the key.
Getting data scopes right is the hard part of multiplayer agents: set data boundaries in Governance before you invite the whole team.
