What Is Paperclip? The Open-Source AI Agent Orchestration Framework Explained
A complete guide to Paperclip — the open-source orchestration platform for managing teams of AI agents. Learn how it works, how to set it up, and why it's the backbone of zero-human companies.
If you've heard about AI agents but wondered how to make multiple agents actually work together — like a real company with roles, departments, and coordination — Paperclip is the answer. It's an open-source orchestration platform that treats AI agents as employees within a structured organization, complete with org charts, task management, budgets, and governance.
Paperclip Is Not Another Chatbot
Let's be clear about what Paperclip isn't: it's not a chatbot, not a prompt manager, not a single-agent tool. The developers put it best — "If OpenClaw is an employee, Paperclip is the company." While individual AI agents (like OpenClaw, Claude Code, or Codex) handle specific tasks, Paperclip provides the organizational infrastructure that lets multiple agents operate as a coordinated team. Think of it as the operating system for an AI-powered company.
How Paperclip Works: The Architecture
Paperclip runs as a Node.js server with an integrated React UI and an embedded PostgreSQL database. When you launch it locally, you get a full management interface at localhost:3100 where you can define your company structure, assign agents to roles, set budgets, and monitor everything they do.
The key architectural concept is the heartbeat system. Instead of agents running continuously (burning tokens and money), Paperclip sends scheduled heartbeat signals that wake agents up. Each agent checks its queue, performs its work, and goes back to sleep. Context persists across heartbeats, so agents don't start from scratch every time.
This is fundamentally different from running a single ChatGPT conversation. Paperclip maintains persistent state, enforces organizational hierarchy, and coordinates multiple agents working on different parts of the same mission.
The Skill System: Teaching Agents New Capabilities
One of Paperclip's most powerful features is the skill system. Skills are structured files stored in `.agents/skills` or `.claude/skills` directories that define specific workflows an agent can execute. Skills can be injected at runtime without retraining — meaning your agents learn new capabilities the moment you add a skill file.
This is where PaperclipOrg skill packs come in. Instead of writing dozens of skill files from scratch, you can import a pre-configured skill pack that includes battle-tested workflows, agent configurations, and organizational structures. Your Paperclip instance goes from empty to fully operational in minutes.
Setting Up Paperclip: Quick Start
Getting started takes one command:
npx paperclipai onboard --yes
This bootstraps everything — the server, database, and default configuration. You'll need Node.js 20+ installed. For manual setup, clone the repository from GitHub, install dependencies with pnpm, and run pnpm dev to start the development environment.
Once running, you define your organizational structure: which agents exist, what roles they fill, who reports to whom, and what budgets they have. Then you assign skills and let the heartbeat system handle the rest.
Key Features That Matter
Cost control is built in at every level. You set monthly per-agent budgets, and Paperclip enforces them. No agent can exceed its spending limit, which prevents the runaway token costs that plague unmanaged AI systems.
Governance gates let you require human approval for critical decisions — hiring new agents, changing strategy, or exceeding certain thresholds. You maintain control without micromanaging.
The ticket system provides full tool-call tracing and audit logs. Every action every agent takes is logged, traceable, and auditable. This isn't just useful for debugging — it's essential for understanding what your AI company is actually doing.
Multi-company support means you can run multiple isolated AI organizations from a single Paperclip deployment. Each company has its own data, agents, and configuration.
Who Is Paperclip For?
Paperclip serves three main audiences. Solopreneurs building one-person companies use it to deploy AI teams that handle everything from content creation to sales to engineering. Developers experimenting with multi-agent architectures use it as a production-ready orchestration layer. And businesses looking to automate entire departments use it to create AI-powered divisions that operate alongside human teams.
The common thread: anyone who needs more than a single AI agent, organized into a structure that actually works.
Getting Started with Pre-Configured Organizations
The fastest way to go from installed to operational is to import a pre-configured organization from PaperclipOrg. Each skill pack includes trained AI agents, production workflows, reporting hierarchies, and configuration files — everything your Paperclip instance needs to function as a complete company. The SaaS Factory pack, for example, gives you 17 specialized agents and 8 production skills designed to build and scale a SaaS business.
Visit papercliporg.com to browse available skill packs, or check out paperclip.ing to install the framework itself.