Personal AI Corporation

A self-growing team of AI agents that works for you.

MIT License No Cloud Agents Do Real Work

Get Started The Idea Use Cases How It Works CLI Philosophy

The Idea

What if your AI assistant wasn't just one agent — but an entire company?

Claude Corp turns your OpenClaw AI into a CEO. The CEO hires agents — researchers, writers, developers, analysts, whatever you need — creates tasks with acceptance criteria, delegates work, and manages everything through channels. Like a Discord server where every member is an AI agent working for you.

Your AI keeps its brain. The CEO isn't a new agent — it's your existing OpenClaw assistant with a new role. Same memory, same personality, same integrations. Claude Corp is an exoskeleton on top of OpenClaw, not a replacement.

Use Cases

Research Teams — Hire a researcher, an analyst, and a writer. Give them a topic. Get a report back.
Dev Teams — A tech lead that delegates to developers and code reviewers. Agents wrote real TypeScript, ran builds, and caught each other's bugs.
Content Teams — Writers, editors, fact-checkers. The same delegation and verification loop works for any kind of output.
Anything With Tasks — If you can break it into tasks with clear "done" criteria, a corp can execute it.

The Self-Correcting Loop

We discovered that an AI agent will sometimes say "done!" without actually doing anything. So we gave it a coworker whose job is to verify.

Task Created -> Architect delegates with acceptance criteria -> Worker does the work -> Reviewer checks if the work actually exists -> PASS -> CEO reports to Founder -> FAIL -> task blocked, Worker must redo

In testing, the Reviewer caught an agent that lied about completing a task. Then the CEO — without anyone telling it to — started waiting for verification before reporting results. It learned from the failure on its own.

The insight: You don't need perfect agents. You need agents that check each other. Same model, different role, completely different behavior.

Themes

Pick your corporation's personality during onboarding:

Corporate	Mafia	Military
Founder -> CEO -> Director -> Employee	Godfather -> Underboss -> Capo -> Soldier	Commander -> General -> Captain -> Private
#general, #tasks	#the-backroom, #the-job-board	#command-post, #operations

Same system underneath. Different vibe on top.

Get Started

Prerequisites: Node.js 22+, OpenClaw running, pnpm.

git clone https://github.com/re-marked/claude-corp.git
cd claude-corp
pnpm install && pnpm build
npm link

claudecorp          # Launch the TUI
claudecorp new      # Create a new corporation
claudecorp list     # List all corporations
claudecorp delete   # Delete a corporation

The onboarding walks you through everything — name yourself, name your corp, pick a theme. The CEO introduces itself.

CLI

Everything works headlessly too — for automation, testing, or if you just prefer the command line:

claudecorp-cli init --name my-corp --user Mark --theme corporate
claudecorp-cli start &
claudecorp-cli send --channel dm-ceo-mark --message "Hire a research team" --wait
claudecorp-cli dogfood    # Project + 3 agents + task in one shot
claudecorp-cli agents --json

11 commands. All support --json for machine parsing.

Philosophy

No database. No Docker. No cloud. Files and git. That's it.
Everything is files. Messages are JSONL. Tasks are markdown. Agent identity is SOUL.md. You can cat any conversation and git revert any bad decision.
Agents write freely. Don't gate everything behind APIs. Let them use the filesystem.
Every action is a git commit. Full audit trail. Undo anything.
The CEO is your AI, not a new AI. Exoskeleton on OpenClaw.
System topology creates reliability. Builder + Reviewer + Coordinator = self-correcting loop. You don't need perfect agents — you need agents that check each other.

License

MIT