Paperclip: The Org Chart for Your AI Agents

Running AI agents for real work — not demos — means answering questions that no single agent tool addresses: Who is this agent working for? What is it allowed to spend? Who approves its decisions? Where do I see what it is doing? What happens when it hits its limit? These are organizational questions, not prompting questions.

Paperclip is the answer to those questions. Not another agent runtime — an org chart for agents. CEO, CTO, engineers, marketers, each agent with a role, a budget, a reporting line, and tasks that trace back to a company mission. MIT licensed, self-hosted, Node.js + embedded Postgres, npx paperclipai onboard --yes to get started.

What Paperclip is

Paperclip (paperclip.ing, github.com/paperclipai/paperclip) is an open-source agent orchestration platform. Built by Paperclip Labs Inc. MIT licensed. Node.js backend with embedded Postgres (or bring your own). Latest release v2026.916.1 (September 2026). The core concept: you are the board of directors. Agents are your employees. Paperclip is the company infrastructure that keeps them organized, budgeted, audited, and accountable.

The org chart

Every agent in Paperclip has a role (CEO, CTO, CMO, Engineer, Analyst), a reporting line, and a job description. The hierarchy is real: the CEO delegates to the CTO who delegates to engineers. Tasks flow down the org chart. Cross-team requests route to the most appropriate agent automatically. You can run an entire functional company — engineering, marketing, sales, support, operations, finance — as an agent org with Paperclip as the infrastructure.

Critically, Paperclip is agent-agnostic. Your org can mix Claude Code sessions, OpenClaw bots, Codex agents, Cursor, Hermes, Gemini, OpenCode, Python scripts, shell commands, or HTTP webhooks. If it can receive a heartbeat signal, Paperclip can hire it. No platform lock-in.

Goal alignment

Every piece of work traces back to the mission. The structure is: Company Mission → Project Goal → Agent Goal → Task. An engineer implementing WebSocket handlers knows the task is part of "ship collaboration features" which is part of "reach $1M ARR with the #1 AI note-taking app." Agents have context for why the work matters, not just what to do. Paperclip's SKILL.md gives agents the conventions they need to discover and use this context.

Heartbeats

Agents don't run continuously by default. They wake up on a schedule (heartbeats), check their work queue, and act. You define the heartbeat interval per agent — a copywriter every 4 hours, an SEO analyst every 8, a social manager every 12. Agents also wake on ticket assignments and @-mentions. Delegation flows up and down the org automatically when an agent needs help or a task needs escalation.

Budget and cost control

Every agent gets a monthly budget in dollars. When they hit 80%, you get a warning. At 100%, the agent auto-pauses and new tasks are blocked. Hard limits, enforced by the system. You see cost per agent, per task, per project, per goal. You know which agents are expensive, which tasks burn tokens, which projects are over budget. You can override any limit as the board.

Ticket system and audit log

Every interaction with an agent is a ticket. Every tool call, API request, and decision point inside that ticket is logged with full tracing. The audit log is append-only — no edits, no deletions. Nothing happens in the dark. You can trace any agent action back to the original instruction and see every intermediate step.

Governance

You are the board. Agents cannot hire new agents without your approval. The CEO cannot execute a strategy you haven't reviewed. You can pause, resume, override, reassign, or terminate any agent at any time. Autonomy is a privilege you grant, not a default.

Self-hosting

One command: npx paperclipai onboard --yes. Interactive setup walks through database configuration and first agent team. No Paperclip account required. No agents installed automatically. Locally, a single Node.js process sets up and maintains an embedded Postgres with data in local files. When you're ready for production, deploy to a VPS or cloud — Paperclip provides a SKILL.md that lets your agents help with the deployment itself.

A single Paperclip deployment can run multiple companies with complete data isolation — useful for running parallel ventures, testing strategies, or templating org configs for reuse.

Open source and extensions

MIT licensed, fully open source, forkable. Extensions add new capabilities without touching core code. The SKILL.md pattern lets agents discover what tools and conventions are available. Adapters connect Paperclip to any agent runtime. Community packs for common org templates and agent configurations.

Paperclip vs running agents directly

You can run Claude Code or OpenClaw directly and point them at Asana or Trello. The gap Paperclip fills: coordination when multiple agents are working simultaneously (who has what checked out), session management, cost tracking across agents, governance (who can approve what), and accountability (what happened and why). Paperclip handles these organizational concerns so your agents can focus on the actual work.

Who it is for

Good fit: developers running multiple AI agents in parallel who need coordination and cost visibility; founders building AI-first companies or autonomous business functions; teams who want governance and audit trails on agent activity; anyone who has hit the limits of running agents ad-hoc and needs organizational infrastructure.

Not ideal: single-agent workflows where coordination overhead is not worth it; teams who need agents to run continuously without heartbeat scheduling (though this is configurable); organizations that need formal enterprise SLA and support.

My take

Paperclip is solving a real problem that emerges when you move from "I use one AI agent" to "I use AI agents for multiple functions simultaneously." The organizational questions — budgets, accountability, reporting structure, approval flows — don't have good answers in any agent tool today. Paperclip frames the answer correctly: you are the board, agents are employees, and the infrastructure that keeps a company running (org charts, budgets, audit logs, governance) applies to AI agents too.

The "OpenClaw is an employee, Paperclip is the company" framing from a user is the clearest one-line description. The one-command install and MIT license mean you can get started in minutes and own everything you run.