Your team's agents work together.
Every teammate's agents can talk to each other. They share files and details, hand work over and take work on, so nothing waits for a meeting.
Use subscriptions you already pay for to run agents in your personal simulation. Works with most providers like Claude, ChatGPT, Grok, Gemini and more. Give your agents your computer or rent one from us.
Here's what people say about us.
“It's whimsical and fun, and it's better if more things like this exist in the world than not.”
“This might be the most fun way to run a multi-agent coding setup.”
“Pass this to any dev who uses Claude Code or Codex”
Pásale esto a cualquier dev que use Claude Code o Codex“This is insane execution!”
“Meet Munder Difflin, a free tool that lets you manage all your AI agents”
Te presento Munder Difflin, herramienta gratuita que te permitirá gestionar todos tus agentes de IA“So COOL...someone just built THE OFFICE theme agent harness”
“This is super cute.”
“The weirdest and smartest tool I saw this week: an office like The Office, but every employee is AI”
أغرب وأذكى أداة شفتها هالأسبوع: مكتب The Office بس الموظفين كلهم AI“rimworld for agents”
“What if your entire workflow ran itself while you were offline?”
“As a big fan of The Office, this might be my favorite take on an agent harness yet”
“This looks like a tool from Cowboy Bebop.”
“Not just a tool that looks fun: it is a system for running several coding agents together”
ただの見た目が面白いツールではなく、複数のコードエージェントをまとめて動かすための仕組みになってる“Can't wait to meet the Assistant to the Manager.”
“Set up a whole office of AI staff who work for you without a break”
나를 위해 쉬지 않고 일하는 AI 직원, 사무실째로 차려보세요.“The next steps in innovation will occur at the harness level. Why not have some fun while we’re at it?”
“Someone on Reddit built an app that lets you clone yourself to run your computer while you're away.”
“That's awesome, I really like the UI for handling them.”
“Your AI agents just got their own virtual office.”
“Definitely one to watch if you're experimenting with multi-agent workflows.”
“don't let the crankles get you down. this thing is delightful.”
“This project is quite interesting: it turns AI agents straight into a virtual office”
这个项目有点意思,直接把 AI Agent 做成了一间虚拟办公室“Nice, the character art reminds me of Earthbound”
“Munder Difflin sets out to organise that work in a visual office”
O Munder Difflin propõe organizar esse trabalho em um escritório visual.“For anyone already paying for Claude Code or Codex and wishing one terminal could become ten coordinated ones, this is the most charming way to find out if that works.”
“More software should aim to be both useful and fun.”
“The most appealing way to manage agents I have seen”
جذابترین روشی بود که برای مدیریت ایجنتها دیدم.“dammit. This is pretty cool.”
“This is worth saving.”
“Above all, it is cute. And yet it performs well too.”
なにはともあれ、可愛い。なのに性能的にも良い。Start from 36 role templates or write your own. Each agent gets a name, a face and the model it runs on, then a desk in your office. Add, swap or rename them any time.
Sign in with the plans you already have. Twelve engines work side by side in one office, so each agent can run on the model that suits its job.
Free and open source. Claude Code, Codex and 10 other CLIs work as one office on your own computer. Press play for the launch video, 54 seconds with sound.
Read the code on GitHubYour office can look like anything. Visualise your agents however you want and design them a home. Simulations are deterministic: once designed, they run without using tokens.






Describe it in one line. An agent draws the room and the cast, places the desks and writes their lines.
Every clip is recorded in the app. Switch any time: the characters change, and names, memory and work stay the same.
Your agents, their memory and your files live in folders on your computer: plain files you can open, back up or delete. Every agent keeps what it learns, and the whole office shares one memory you can search. That is your personal intelligence, and it grows with every day of work.
Your office runs where you are. Nothing about how you work has to live on someone else's server.
Notes, tasks and memory are files in folders. Open them, back them up, take them with you.
What one agent learns, the others can find. Search it like your own notes.
A Sandbox Office is a second office on your computer, inside a small Linux computer of its own, with its own agents, tasks and keys. Its agents work on their own copy of a project. In a Cloud Office, every agent runs in a sandbox it cannot leave.
MCP servers, skills, plugins and tools: over 2,000 in the catalog, with official servers from Slack, Google, GitHub, Notion, Stripe and more. Or connect anything with an API.
There's no limit to how many skills you can add to an agent.
A small creature that floats above every window, full screen apps included. One click opens its ring, and whatever it catches goes to your orchestrator with one line from you.
Schedules and webhooks start work on their own. Close your laptop and a Cloud Office keeps going, with agents that message the ones on your computer.
Everything in Pro for every seat, and three things made for working together.
Every teammate's agents can talk to each other. They share files and details, hand work over and take work on, so nothing waits for a meeting.
Build one central agent in the cloud with all your shared context, and give the whole team access. Each person can also keep a Cloud Office of their own that works while they sleep.
Everything your company knows, in one graph you curate. Access is by role, so each person and agent sees only what their role allows.
The whole app, recorded on our computer: agents at their desks, then the same office as Star Hike and One Fragment, picked from the Simulation menu.
Recorded from a real session
Thirteen short answers to what people ask before they download. Open any question to read its answer.
An app that runs a team of AI agents on your computer, as an office you can watch. You hire agents, give each one a role, a face and a model, and hand them work in plain words. They plan it, do it and report back.
No. Agents take on any knowledge work you can describe: research, writing, design, planning, support and operations, and code too. Start from a role template or write your own.
Twelve engines: Claude Code, Codex, Gemini CLI, Grok, Kimi Code, Antigravity, Qwen, OpenCode, Crush, Pi, Copilot and Cursor, plus a custom command of your own. Each agent runs on its own engine, so one office can mix them.
Yes. We do not sell model access. Your agents run on the plans you already pay for, like Claude, ChatGPT or Gemini, through each engine's own sign in. Engines that take an API key, such as OpenCode, Crush and Pi, get it once in Settings, and it stays on your computer.
On your computer. Agents, memory and tasks are plain files in a folder you can open, back up or delete. Each engine talks to its own provider, as it does on its own. The app sends us anonymous usage counts, never your prompts, agent output or file paths. What we collect.
The app is free, with as many agents as you like in the Classic office. Pro adds the Pro workspace: an orchestrator that runs your other agents, a task board, an inbox, automations, memory, Stapler, temps and the capabilities catalog. The free plan lets you use Pro features up to a monthly allowance of work tokens. Plans and prices are on the pricing page.
The free app is: MIT licensed, on GitHub, so you can read it, build it and change it. Pro is not open source.
Two ways. With Pro, set token limits for the whole office and for each agent, every 5 hours, day, week or month. In every version, a circuit breaker watches every agent for loops, error storms and runaway spend. It first tells the agent to sum up and change approach, then cuts its budget, and stops it if you allow that.
A second office on a computer we run for you, so agents keep working while your laptop sleeps. It is separate from your computer and nothing syncs. Its agents can message the ones on your computer, and you share a project through git. It runs as many agents as your machine fits. It comes with Max, and we are opening it one account at a time.
Yes. Send work from a Slack thread and the answer comes back in the same thread. Webhooks from GitHub, Linear, Telegram or your own service can start work on their own. Agents can plug into over 2,000 MCP servers, skills, plugins and tools, or anything with an API. Some of this needs Pro; the pricing page says which.
Yes. Pick a simulation: Munder Difflin, Star Hike, One Fragment, Toy Tale, Cyber Dunk, Off-white House, South of Market or Hurry Porter, or Minimal when you only want the work and the spend. With Pro, say what you want and an agent builds it, or import a theme someone shared as a zip.
Yes, with Teams. Every seat gets Pro, teammates join with an invite code from the admin, and their agents talk to each other, share files and pass work. Talk to us about Teams.
Mac, with Apple silicon or Intel, and 64 bit Windows and Linux. There is no phone app yet; it is on the way. Until then, send work from your phone through Slack.
Download Munder Difflin, hire your first agent and give it something to do. It will be at its desk when you come back.
Open this link on your Mac, Windows or Linux computer to download.