engineering / agent

Munder Difflin

A desktop app that runs terminal coding agents such as Claude Code and OpenAI Codex side by side, with shared memory and an orchestrator agent you talk to.

At a glance

Purpose
A desktop app that runs terminal coding agents such as Claude Code and OpenAI Codex side by side, with shared memory and an orchestrator agent you talk to.
Runs in
Local agent with shell
Operating systems
macOS, Linux, Windows
Needs
Runtime: Not stated; Needs internet: Not stated; Needs accounts or API keys beyond your agent: Not stated
Files and data
Reads files: Yes; Writes files: Yes; Sends data to other services: Yes
Tool type
Agent
Delivery
From GitHub
Version
Latest on the main branch
Access
Code on GitHub
Compatibility
According to the README, Munder Difflin runs on macOS, Windows and Linux. It lists these agent CLIs: Claude Code, Antigravity (Gemini), OpenAI Codex, xAI Grok, Kimi Code, Gemini CLI, Qwen, OpenCode, Crush, Pi, GitHub Copilot CLI and Cursor, plus any custom command. It also lists bring-your-own keys and local models through Ollama, LM Studio or vLLM. The README names Slack and inbound webhooks for GitHub, Linear and Telegram as ways to send work in.
External costs
According to the README, Munder Difflin is free and open source and works with the subscriptions you already have for the agent CLIs it runs, within their hourly limits. The README also lists bring-your-own API keys and local models. Your coding agent or model provider may charge for usage. The README describes a paid Pro plan that adds Stapler and a newer sidebar, and it mentions a launch offer for the annual plan with the price adjusted for purchasing power. Check the README for current prices.
Permissions
The README says agent CLIs run as local processes using existing subscriptions, supplied API keys or local endpoints. The README describes per-agent autonomy settings and escalation of spend, scope changes and destructive operations. The README lists worker worktrees, file editing, Git operations, scheduled missions, heartbeat processing, a CI watcher and a hook server. The README describes Slack-triggered workers that reply in-thread and inbound webhooks. The README says the app can install missing CLIs and download, restart and install application updates. The README's source route installs npm dependencies and rebuilds node-pty. The README says macOS dictation needs Microphone and Accessibility permissions and describes recording microphone and meeting audio.
Usage terms
MIT License, as stated in the repository. See its LICENSE file for the exact terms. The README says MIT covers the source code only. The README says bundled LimeZu Modern Interiors pixel art is separately licensed under its Complete Version licence, requires retained LimeZu credit and is carved out in LICENSE-ASSETS.
Interface
desktop

Code on GitHub

HarnessMD/munder-difflin

MIT

TypeScript8,700 stars1,191 forksLast pushed 12 Oct 2026 UTC

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

Topics: agent-harness agent-orchestration agents ai-agents autonomous-agents claude-code codex coding-agents

Public GitHub data fetched · 0 days old

View on GitHub ↗

What it does

The README describes Munder Difflin as an Electron desktop app wrapping terminal agent CLIs such as Claude Code, Codex, Gemini CLI and OpenCode. The README says each runs as a process in its own pseudo-terminal using existing subscriptions and their hourly limits.

The README says agents appear as avatars on a 2D office floor. The README describes a shared hive of plain files containing memory, mailboxes, a blackboard and an event log. The README describes Michael, also called the GOD agent, as an orchestrator that assigns work, routes messages and escalates spend, scope changes and destructive operations.

The README lists optional Git worktrees, scheduled missions, a Monaco editor, a skills browser, Slack and webhook inputs, dictation and meeting transcription. The README says Stapler and the newer sidebar belong to a paid Pro plan.

Who it is for

Developers who already use terminal coding agent CLIs and want to run several of them together in one desktop app.

Example tasks

Add an agent by choosing a CLI, a model and an autonomy level.
Brief the orchestrator agent and let it route work between agents.
Click an agent's desk to read its terminal live and type into it.
Set how far each agent may go before spend, scope or destructive operations come back to you.
Search the shared agent memory from the app.

Getting started

According to the README, most people download the app from the project's download page or its GitHub releases page, install it and follow the onboarding wizard. You still need at least one agent CLI on your machine. A missing CLI shows as a card in the agent's terminal, and the README says Install runs its command. To run from source, clone the repository, then run npm install and npm run dev. Source builds need Node.js 18+, npm and a C/C++ toolchain for node-pty. The README has the exact commands.

Limitations

The README labels the project pre-release and lists macOS, Windows and Linux. The README requires at least one supported agent CLI and describes use of existing subscriptions within their hourly limits. The README lists Node.js 18+, npm and a C/C++ toolchain for source builds. The README says Stapler and the newer sidebar require Pro. The README describes global dictation on macOS, Windows and Linux X11, with Wayland unsupported for that workflow. The README says the bundled transcription model in 0.5.3 is English-only; on macOS 26, dictation uses Apple's recogniser and follows the Mac's language, with English reported as the tested language. The README lists an unresolved Kimi scroll and queue bug in 0.5.3. The repository can change at any time; check its README before you install.

Data practices

The README says agent memory and mailboxes live in a local Git repository of plain files. The README describes local transcription with a bundled Whisper model by default, Apple's on-device recogniser on supported macOS versions and Groq when the user supplies a key. The README describes a write-only API-key broker. Data sent by agent CLIs depends on their configured providers.

The README's Telemetry section says official builds send anonymous usage events such as app opening, agent spawning and feature use. The README lists a Settings toggle, DO_NOT_TRACK and building from source as opt-out routes, and says forks compile without the telemetry key. The README points to TELEMETRY.md for the event list and further details.

Community experience

Not rated yet

Sign in to share your experience.

Comments 0

No comments yet.

Sign in to comment

Report a listing concern