Munder Difflin: Run an Office of Your AI Clones
Munder Difflin is an open-source agent harness that turns your existing CLI-based AI agent into a 24/7 clone of you, running on your own laptop. It wraps the agent CLI you already use (e.g., Claude Code, Codex, Grok, Kimi) and enables multiple clones to communicate, hand off work, and unblock each other — all with end-to-end encryption and local-first architecture.
How It Works
- Install the harness — One download, wraps your existing agent CLI, runs on your laptop. Your keys, code, and subscription stay local.
- It becomes you — Captures your workflow, tooling, and knowledge. Every clone starts with shared memory.
- Your office works around the clock — Clones message each other when blocked, hand off context, and unblock each other on your machine.
Key Features
- Local-first — Each clone is a node on the owner's laptop. Code and keys never leave the machine (everything runs at
127.0.0.1). - End-to-end encrypted — Clone-to-clone messages are encrypted on your node and decrypted only on your teammate's. The provider (Munder Difflin) cannot read them.
- Isolated worktrees — Each agent runs in its own isolated
gitworktree to prevent conflicts. - MemPalace — Shared memory system that stores your workflow and knowledge, inherited by new clones.
- Shared knowledge with rules — Org-level context provisioned once, versioned, and inherited; personal vs. shared is strictly separated.
- Open source — MIT licensed; node, protocol, and crypto all on GitHub for audit.
What Your Clones Can Do
Each teammate gets a clone that can do their job — since everything a computer does is reachable from the command line:
- Developer — Review PRs, fix bugs, ship small features, babysit CI, keep docs honest.
- Designer — Audit screens against design system, export assets, draft specs.
- Product manager — Write specs, triage issues, sync boards/docs, prep standup summaries.
- Sales & GTM — Draft outreach, prep call briefs, keep CRM honest, chase follow-ups.
- Everyone else — Reports, spreadsheets, scheduling — anything scriptable.
Security Details
The encrypted message log shows exactly what an intermediary sees — only encrypted blobs with cipher info:
03:12:07 · jim-clone → pam-clone · 1.2 KB · X25519 / AES-256-GCM nQf4x9Uc2mL8…J1sKw0Yd7Rz3TgHveA5oP6iB4tCkXhSMDrEyWuNa8lF2mQ==
Clones escalate only decisions that genuinely need a human — you check in occasionally, answer, and it keeps moving. You stay the boss.
Munder Difflin is for teams already using CLI-based AI agents who want asynchronous, private collaboration — not a shared bot. It’s particularly useful for distributed teams that need answers at 3 AM without waking anyone.
📖 Read the full source: HN AI Agents
👀 See Also

HolyCode: Docker Container for Persistent Claude AI Coding Environments
HolyCode is a Docker container that maintains AI coding environment state across machine switches and rebuilds. It includes 30+ preinstalled tools, browser automation with Chromium + xvfb + Playwright, and preserves context in ./data/opencode.

Sgai: Goal-Driven Multi-Agent Software Development Tool
Sgai is an open-source Go tool that coordinates AI agents to execute software goals defined in GOAL.md files. It decomposes goals into DAG workflows, runs tests for completion gates, and operates locally with a web dashboard for monitoring.

graphify-ts: Local MCP server cuts Claude Code PR review tokens from 63K to 8.7K
graphify-ts builds a local knowledge graph of your codebase using tree-sitter AST + Louvain communities + BM25 + optional ONNX rerank, exposing it via MCP stdio. In production tests, it reduced input tokens by 2.6x and latency by 2.8x for code queries, and cut PR review prompts from 63K to 8.7K tokens.

Brain-MCP Developer Documents Tools for Claude AI Instead of Humans
A developer maintaining the Brain-MCP server added a 'For AI Assistants' section to documentation with behavioral instructions, resulting in Claude using tools more intelligently and proactively injecting context when topics change.