chaitanyagiri/munder-difflin

local multi-agent harness

What it solves

Munder Difflin is a desktop application that transforms individual terminal-based AI coding agents (such as Claude Code, GitHub Copilot CLI, and others) into a coordinated, multi-agent team. It solves the problem of managing multiple isolated AI agents by providing a shared memory layer, a communication system, and a central orchestrator to route tasks and manage workflows.

How it works

The system wraps real terminal processes (using node-pty) and organizes them into a "hive mind."

  1. Orchestration: A central "GOD agent" acts as a supervisor that you interact with to assign tasks, route work between agents, and handle escalations.
  2. Communication: Agents collaborate via a local git repository of plain files, using an inbox/outbox system for messaging.
  3. Memory: A markdown-first long-term memory layer with a semantic recall index allows agents to remember information across sessions.
  4. Visualization: The entire process is visualized as a 2D office floor using Pixi.js, where agents are represented as avatars that move and send envelopes to one another.

Who it’s for

Developers who already use terminal-based AI coding agents and want to move from single-agent interactions to an autonomous, multi-agent workforce with shared memory and a built-in IDE for reviewing changes.

Highlights

  • Multi-Engine Support: Compatible with Claude Code, Antigravity, OpenAI Codex, OpenCode, Crush, pi.dev, and GitHub Copilot CLI, including support for local LLMs via Ollama or LM Studio.
  • Built-in Monaco IDE: A self-hosted editor with a file tree and side-by-side git diffs to review agent-driven changes.
  • Realtime Voice Control: A low-latency voice channel to the GOD orchestrator using the OpenAI Realtime API.
  • Observability & Control: Includes a cost/runaway circuit breaker, token telemetry, and a human-in-the-loop approval queue for critical operations.
  • Integration Ecosystem: Features a task kanban board, GitHub issue ingestion, CI status monitoring, and the ability to spawn ephemeral workers from Slack requests.