munder-difflin
an open-source alternative to the dots, bots and muses of the world, run an office of claude code/codex like agents on your laptop, sandboxes or anywhere, uses
What is munder-difflin?
an open-source alternative to the dots, bots and muses of the world, run an office of claude code/codex like agents on your laptop, sandboxes or anywhere, uses your existing subscriptions
Overview of munder-difflin
What it is
Munder Difflin is a free, open-source desktop application that wraps multiple terminal-based AI coding agents (such as Claude Code, OpenAI Codex, xAI Grok, Gemini CLI, and others) into a coordinated team. Each agent runs as a real process in its own pseudo-terminal, with its own memory, mailbox, and a visual avatar on a 2D office floor. A central agent named Michael acts as the orchestrator, routing work between agents and escalating only critical items to the user. The application is built with Electron, React, TypeScript, Pixi.js, xterm.js, and node-pty.
Key features
- Wraps existing terminal-agent CLIs (Claude Code, Antigravity/Gemini, OpenAI Codex, xAI Grok, Kimi Code, Gemini CLI, Qwen, OpenCode, Crush, pi.dev, GitHub Copilot CLI, Cursor) and supports custom commands, bring-your-own keys, and local models via Ollama, LM Studio, or vLLM.
- Each agent has long-term markdown memory with a semantic recall index, allowing it to remember across sessions and recall information in milliseconds.
- Agents collaborate through a shared hive (a local git repo of plain files) with outbox/inbox messaging, coordinated by the GOD agent (Michael).
- Visualizes agents as avatars on a Pixi.js office floor; they walk to stations when working and send envelopes when messaging.
- Includes dictation (hold Option on Mac, Control+Alt+Space on Windows/Linux) and meeting transcription (Stapler feature in Pro) that runs locally using a built-in Whisper model.
- Provides agent cards showing status, engine, model, context gauge, current ticket and action, with a right-click menu. Pro adds a sidebar with Tasks, Inbox, Automations, Memory, Capabilities, and agent grouping.
Use cases
- Running a team of AI coding agents that work autonomously on different tasks while the user monitors and intervenes only when necessary.
- Automating software development workflows by assigning agents to specific roles (e.g., code review, testing, documentation) with long-term memory.
- Using dictation and meeting transcription features to capture notes and meeting minutes locally without cloud services.
- Experimenting with multiple AI models and providers in a single coordinated environment.
Getting started
See the repository README for installation instructions. The latest release (0.5.3) is available for macOS, Windows, and Linux from the GitHub releases page. macOS builds are signed and notarized; building from source is not required.
License and pricing
The software is licensed under the MIT License. It is open source and free to use. A Pro annual plan is available for $150 USD (adjusted for purchasing power, as low as $100 per year) which unlocks additional features such as the Stapler meeting transcription, sidebar, and agent grouping.
Based on the project's public README and repository metadata.
