munder-difflin is a desktop application that runs a team of AI agents as an office on your own computer, driven by the CLI subscriptions you already pay for. Agents, memory, and tasks are stored as plain files you control.
Project overview
It targets a concrete gap: people pay for AI CLI subscriptions but previously had no way to coordinate a team of agents on their own computer using those existing plans, and this project runs each agent through the real, signed-in CLI rather than selling model access.
Project type
AI Agent
Use cases
Coding & Development · Automation
Deployment
Refer to project documentation
License
MIT
Best for
Developers who already pay for AI CLI subscriptions and want to coordinate a team of agents locally on their own computer
Users who want agents, memory, and tasks stored as plain files under their own control
Key capabilities
A desktop app that runs a team of AI agents as an office you can watch; you hire agents with roles, faces, and models, and hand them work in plain words.
Limitations and risks
Releases ship as source code, so you must build the desktop app yourself.
Setting up and running the project requires coding ability.
Each agent runs a real CLI signed in with your own plan; the project does not provide model access, so paid service usage is an external dependency of running agents.
The project's data boundary allows optional external interaction, so a fully local-only end-to-end boundary is not established.
Getting started
Setup is moderately difficult and requires coding ability. First success path: git clone https://github.com/HarnessMD/munder-difflin.git, cd munder-difflin, nvm use && npm install, npm run dev.
Evidence and sources
GitHub project description: Coolest app to use agents, local, multi provider and open source grok bot alternative. Build custom worlds for your claude code and codex like agents
README: Munder Difflin is a desktop app that runs a team of AI agents 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 wo…
README: Twelve engines work side by side in one office, plus any command of your own. Each agent picks its own, so one office can mix them.
README: Michael runs the floor. He handles routine requests himself and puts anything that needs you (spend, deletions, a change of scope) in Ask me.
README: Slack. Send work from a thread and the answer comes back in the same thread.