HarnessMD/munder-difflin

What it solves

Munder Difflin acts as a central command center for managing multiple AI coding agents. It solves the problem of juggling disparate terminal-based AI tools by providing a unified interface to coordinate, monitor, and manage a team of agents that work in parallel on your machine.

How it works

The project wraps existing terminal-based AI agent CLIs (such as Claude Code, GitHub Copilot, and others) into a "hive" architecture. A primary "orchestrator" agent (Michael) manages the workflow, assigning tasks and routing messages between specialized agents. Each agent runs as a real process in its own terminal, maintains its own long-term memory, and is visualized as an avatar on a 2D office floor. The system uses a semantic memory layer to ensure agents recall past interactions across sessions.

Who it’s for

It is designed for developers and power users who already rely on terminal-based AI coding assistants and want to scale their productivity by running multiple agents simultaneously with centralized oversight and human-in-the-loop controls.

Highlights

• Unified Agent Management: Orchestrates multiple CLI agents (e.g., Claude, Grok, Gemini, Qwen) within a single desktop application.
• Visual Office Interface: Provides a 2D office floor visualization where agents act as avatars, allowing you to monitor their status and activity in real-time.
• Human-in-the-loop: Includes safety features like a "circuit breaker" and approval gates for critical operations like spending or destructive file changes.
• Persistent Memory: Features a semantic recall index that allows agents to remember project context and previous tasks across different sessions.
• Integrated Tools: Includes built-in dictation, meeting transcription, and a Monaco-based IDE for managing files and git history directly within the app.