Paperclip: Run a Zero-Human Company With AI Agents

Most people running multiple AI agents end up with the same mess: 20 Claude Code tabs open, no idea who's doing what, sessions that die on reboot, and runaway token spend you don't notice until the bill arrives.

Paperclip is built to fix that. It's an open-source Node.js server + React UI that treats AI agents like company employees — with org charts, budgets, goal alignment, governance, and a ticket system. You define the mission, hire agents, set budgets, and the company runs.

Released under MIT license, it hit 14,000 GitHub stars in its first week. Here's what it actually does and how to get started.

---

The Core Idea

Paperclip's tagline: "If OpenClaw is an employee, Paperclip is the company."

That's not just a metaphor. Paperclip models a real organizational structure:

• Goals flow from the company mission down to every task — agents always know why they're doing something, not just what
• Org charts define roles, reporting lines, and delegation — a CEO agent supervises engineering agents, who can delegate to sub-agents
• Budgets cap monthly spend per agent — when they hit the limit, they stop. No runaway loops
• Heartbeats trigger agents on a schedule — recurring jobs (support, social, reports) happen automatically
• Tickets track every task, conversation, and decision — with full tool-call tracing and an immutable audit log
• Governance lets you (the human board) approve hires, override strategy, or pause any agent at any time

It's not a workflow builder or a chatbot. It's company infrastructure for agent teams.

---

Who It's For

Paperclip is explicit about this: if you have one agent, you probably don't need it. If you have twenty — you definitely do.

You're the right fit if:

• You coordinate multiple agents (OpenClaw, Claude Code, Cursor, Codex) toward shared goals and lose track of what's happening
• You run recurring automated jobs and manually kick them off
• You want persistent agent sessions that survive reboots
• You want to audit costs and enforce budgets on AI work
• You're building autonomous systems and want governance without babysitting

---

Quickstart

Requirements: Node.js 20+, pnpm 9.15+

The fastest path:

npx paperclipai onboard --yes

This scaffolds the config, starts the API server at http://localhost:3100, and creates an embedded PostgreSQL database automatically — no separate database setup required.

Or manually:

git clone https://github.com/paperclipai/paperclip.git
cd paperclip
pnpm install
pnpm dev

The React dashboard opens at http://localhost:3100. From there, create your first company, define the mission, and start adding agents.

---

Core Features Explained

Bring Your Own Agent

Paperclip doesn't ship agents. It orchestrates them. Any agent that can receive a heartbeat ping — OpenClaw, Claude Code, Cursor, a custom script — can be "hired" and added to the org chart. You configure their prompt, model, runtime, and budget. Paperclip handles the rest.

Heartbeats

Agents don't run continuously by default. They wake on a schedule (heartbeat), check for assigned work, complete it, and report back. This prevents runaway loops and makes costs predictable. You set the heartbeat interval per agent.

Goal Alignment

Every task carries its full goal ancestry — the agent knows the task's parent project, the project's company goal, and the company's mission. This eliminates context drift where agents optimize for a local task at the expense of the actual objective.

Cost Control

Monthly token budgets per agent. When an agent hits its limit, it stops and notifies management. No surprise bills. Budget enforcement is atomic — checked at task checkout to prevent double-spend.

Multi-Company Support

One Paperclip deployment can run multiple companies with complete data isolation. If you're running several autonomous businesses or projects, a single instance manages all of them from one dashboard.

Mobile-Ready

The React dashboard works on mobile browsers. Monitor agent activity, approve tasks, and check costs from your phone without needing a desktop.

---

What Problems It Actually Solves

Without Paperclip	With Paperclip
20 Claude tabs open, sessions die on reboot	Persistent sessions, ticket-based task tracking
Manually gather context to brief each agent	Context flows from task → project → company goal automatically
Runaway loops burn hundreds of dollars unnoticed	Monthly budgets per agent, atomic enforcement
Recurring jobs require manual kickoff	Heartbeats handle scheduling — agents work without prompting
No audit trail of what agents decided	Full tool-call tracing, immutable audit log

---

Deploying Beyond Local

For solo operators: the FAQ recommends using Tailscale to access Paperclip remotely without exposing it to the internet. This is the cheapest setup — your existing machine runs the server, Tailscale lets you reach the dashboard from anywhere.

For production: Paperclip supports any Postgres instance and standard Node.js deployment. Vercel, Railway, Fly.io, or a VPS all work. The embedded Postgres is fine for local dev; point it at a managed database for production use.

For the GPU compute side — if your agents are doing AI inference or you want them to run models locally — Ampere.sh is worth looking at for affordable ARM-based GPU instances. Significantly cheaper than AWS for AI inference workloads.

---

What's on the Roadmap

Already shipped:

• Plugin system, OpenClaw agent employees, org import/export (companies.sh), Skills Manager, Scheduled Routines, Better Budgeting

Coming soon:

• CEO Chat (natural language control from within the dashboard)
• Artifacts & Deployments
• MAXIMIZER MODE (details not disclosed yet, but it's on the list)
• Multiple Human Users
• Cloud/Sandbox agents (Cursor, e2b)
• Desktop App

The roadmap is moving fast — this project is a week old and already shipping production features.

---

Key Takeaways

• Install: npx paperclipai onboard --yes — embedded Postgres, no setup required
• What it is: Company infrastructure for AI agent teams — org charts, budgets, governance, tickets
• What it's not: A chatbot, workflow builder, or single-agent tool
• Best for: Anyone running 5+ agents with overlapping goals and no structure
• Cost control: Monthly budget caps per agent, atomic enforcement — runaway loops can't happen
• License: MIT — open-source, self-hosted, free
• GitHub: paperclipai/paperclip — 14k stars in week one