Paperclip is AI agent orchestration for builders: org charts, budgets, goals, governance, and audit logs all in one self-hosted MIT-licensed Node.js app.

Managing twenty AI agents across twenty browser tabs is not a workflow. It is a liability. Context disappears on reboot, costs spiral before anyone notices, and no single agent knows how its work connects to the larger mission. Paperclip is an open-source Node.js server and React UI built to solve exactly this problem by providing a complete organizational layer that turns a pile of autonomous bots into a functioning company.

The project describes itself simply: if an AI coding agent is an employee, Paperclip is the company. That framing clarifies what it does and what it does not do. Paperclip is not a new agent runtime, a prompt library, or a drag-and-drop workflow builder. It is the company layer above the agents, defining hierarchies, delegating goals, enforcing budgets, and logging every decision in an immutable audit trail.

What AI Agent Orchestration Actually Looks Like in Paperclip

Spinning up Paperclip takes one command: npx paperclipai onboard --yes. The interactive setup walks through database configuration and first company creation. An embedded PostgreSQL database starts automatically, so no external database setup is required for local use. The stack is Node.js 20 with pnpm, self-hosted, MIT-licensed, and requires no Paperclip account.

Once inside, builders define a company mission, hire agents from any supported runtime, and assign goals that flow down from the company objective to projects to individual tasks. The platform ensures every piece of work carries its full goal ancestry. An agent writing a WebSocket handler knows it is building real-time sync, which ships collaboration features, which serves the mission of reaching one million in monthly recurring revenue. Context does not need to be re-pasted every session.

Agents wake on scheduled heartbeats rather than running continuously by default. This keeps costs predictable. Each agent has a monthly token budget enforced at the infrastructure level: at eighty percent utilization a soft warning fires, at one hundred percent the agent auto-pauses and new tasks are blocked until the board, meaning the human operator, overrides the limit. Atomic task checkout prevents two agents from working on the same ticket simultaneously, eliminating wasted compute from duplicated effort.

AI Agent Orchestration With Full Governance Built In

The governance model places the human operator as board chair. Agents cannot hire new agents without approval. The CEO agent, whichever model fills that role, cannot execute a strategy the board has not reviewed. Any agent can be paused, reassigned, or terminated at any time. Config changes are versioned and can be rolled back. This positions Paperclip as structured governance rather than a fully autonomous black box: autonomy is a privilege the operator grants incrementally, not a default state.

The ticket system ties everything together. Every instruction, every response, every tool call, and every decision gets recorded with full tracing. The audit log is append-only. Nothing happens outside a ticket thread. This matters for builders running multiple companies from a single Paperclip deployment: complete data isolation means one install can manage a content agency, a trading desk, and a software studio as three entirely separate companies with separate budgets and separate audit trails.

ClipMart and the Future of Portable AI Companies

The most forward-looking feature on the roadmap is ClipMart, a marketplace for complete pre-built company templates. The concept is one-click import of an entire org structure: roles, agent configurations, goal hierarchies, and skill definitions bundled into a portable file with secrets scrubbed before sharing. A content marketing agency template, a crypto trading desk, a YouTube factory, or a dev agency could each be downloaded and running in minutes.

Paperclip sits at an interesting intersection. It is neither a competing AI runtime nor a task manager bolted onto an existing tool. It is purpose-built AI agent orchestration infrastructure for the emerging class of solo entrepreneurs and indie builders who are already running multiple agents but doing so without any organizational structure. The project is open source under MIT, available at github.com/paperclipai/paperclip, with documentation at paperclip.ing/docs.

Source: Paperclip on GitHub by the Paperclip team.