Paperclip AI: Complete Guide to Open-Source AI Agent Orchestration (2026)
Paperclip AI: The Complete Guide to Open-Source AI Agent Orchestration
Paperclip AI is an open-source platform for running a company of AI agents. Instead of juggling a dozen chat windows, you give agents job titles, a reporting line, a monthly budget and a ticket queue, and Paperclip wakes them up on a schedule to do the work. It launched in March 2026, passed 30,000 GitHub stars in three weeks and sits at roughly 85,000 stars as of September 2026. This guide explains what it is, how it works, how to install it, what it costs, how it compares to OpenClaw, Hermes and CrewAI, and how a business can actually put it to work.
Neon Digital Media is a Las Vegas automation agency. We build AI workflow automation and n8n automation for local businesses, and Paperclip is now one of the tools we deploy when a client needs more than one agent working together. Everything below is written from that hands-on perspective.
What Is Paperclip AI?
Paperclip (often searched as "Paperclip AI" or "paperclipai") is an MIT-licensed, self-hosted application that manages teams of AI agents the way a company manages employees. Its own tagline is "the open-source app everyone uses to manage agents at work," and the one-line explanation the project uses is memorable: if OpenClaw is an employee, Paperclip is the company.
That framing matters because Paperclip is deliberately not several things people assume it is:
- Not a chatbot. There is no chat window where you talk to "Paperclip." You talk to the agents it manages.
- Not an agent framework. It does not tell you how to build an agent. It hires agents that already exist, such as Claude Code, Codex, Cursor, OpenClaw, Hermes or OpenCode.
- Not a drag-and-drop workflow builder. Tools like n8n, Make and Zapier automate fixed sequences of steps. Paperclip assigns goals and tickets to agents that decide their own steps.
- Not for a single agent. If you run one assistant, the overhead is not worth it. Paperclip earns its keep at five, ten or twenty agents.
What it is: a Node.js server with a React dashboard that stores companies, org charts, goals, tickets, budgets and an immutable audit log in PostgreSQL, and that sends "heartbeats" to agents so they wake up, check their queue, do work and report back. The project calls its four pillars the Agentic Task Manager, the Org Chart for Agents, Agent Employee Training (skills) and the Agentic OS (the infrastructure underneath).
| Fact | Detail (as of September 2026) |
|---|---|
| Official website | paperclip.ing (docs at docs.paperclip.ing) |
| Repository | github.com/paperclipai/paperclip |
| License | MIT, copyright Paperclip Labs, Inc. |
| Launched | Early March 2026 |
| GitHub stars | About 85,000 (15,000 forks) |
| Latest release | v2026.916.1 (September 21, 2026) |
| Stack | Node.js, TypeScript, React, PostgreSQL (embedded locally) |
| Default port | API and dashboard at http://localhost:3100 |
| Cost | Free to self-host. You pay for hosting and LLM usage. |
Who Created Paperclip AI, and Why
Paperclip was built by a pseudonymous developer known as @dotta and is maintained by Paperclip Labs, Inc. The origin story is relatable to anyone who has tried to run several coding agents at once. Dotta was operating an automated trading operation and ended up with more than twenty Claude Code tabs open at the same time. None of them shared context, none of them knew what the others were doing, and when the machine rebooted every bit of state was gone.
Paperclip was the fix: put the agents in an org chart, give each one a job description, route work through tickets so two agents never grab the same task, and persist everything in a database so a reboot means nothing. The launch tweet in March 2026 drew around 2.4 million views, the repository crossed 30,000 stars within three weeks, and by June it had passed 70,000. That growth put it among the fastest-growing open-source projects GitHub has ever recorded and made "zero-human company" a phrase people search for. We cover the repo itself in Paperclip AI on GitHub: the repo explained.
How Paperclip AI Works
Every concept in Paperclip maps to something a manager already understands. Here is the full vocabulary, and how the pieces connect. For a deeper walkthrough read how Paperclip agent orchestration works.
Companies and the org chart
A company is the top-level container. One Paperclip instance can host several companies with complete data isolation, which is why agencies like ours can run one instance per client account. Inside a company you build an org chart: roles, titles and reporting lines. A "CEO" agent can delegate to a "Head of Content" agent, who delegates to two "Writer" agents. Each agent has a job description, a set of permissions, and an adapter that says which runtime actually executes the work.
Goals and tickets
Goals are the mission tree. Every task can trace its ancestry back to a company goal, so an agent always knows why it is doing something. Issues (tickets) are the unit of work. Checkout is atomic: when an agent picks up a ticket, no other agent can pick it up, which is what stopped dotta's twenty tabs from stepping on each other. Tickets carry comments, status, attachments and a full trace of every tool call the agent made.
Heartbeats
A heartbeat is a scheduled wake-up. Paperclip pings an agent on an interval (or when a ticket is assigned), the agent checks its queue, works, and reports back. Because the schedule lives in Paperclip rather than in the agent, an agent that crashes or reboots simply gets its next heartbeat and carries on. This is the mechanism behind the "if it can receive a heartbeat, it's hired" rule.
Budgets and cost tracking
Each agent gets a monthly budget in dollars. Paperclip tracks spend by company, agent, project, provider and model, and it auto-pauses an agent at 100 percent of its budget. For a business owner this is the single most important feature: a runaway agent cannot produce a surprise API bill.
Approvals and board governance
You sit on the board. Hiring a new agent, changing strategy, merging code or spending above a threshold can all be gated behind an approval. You can pause, resume or terminate any agent at any time, and the audit log is append-only, so every decision and every tool call is explained and reviewable.
Routines, secrets, workspaces, skills and plugins
- Routines are cron-style recurring jobs (daily reports, weekly reviews).
- Secrets are encrypted API keys and credentials agents can use without seeing them in plain text.
- Workspaces isolate work in git worktrees and branches, so parallel coding agents do not collide.
- Skills are SKILL.md files that teach an agent how your company does a task.
- Plugins run as out-of-process workers and can add tools, scheduled jobs and dashboard panels. Linear, Slack and Telegram integrations ship this way.
Which AI Agents Can Paperclip Run?
Paperclip is model-agnostic and runtime-agnostic. It supplies the management layer; adapters connect it to the agents that do the work.
| Adapter | What it runs | Typical role |
|---|---|---|
| Claude Code | Anthropic's terminal coding agent | Engineering, data work, content pipelines |
| Codex | OpenAI's coding agent | Engineering |
| Cursor | Cursor agent | Engineering |
| OpenCode / Pi | Open-source harnesses; pair with OpenRouter for any model | General work on cheaper or open models |
| OpenClaw | Personal-assistant gateway with WhatsApp, Telegram, Slack and Discord channels | Customer-facing and ops roles |
| Hermes Agent | Nous Research agent that builds its own skills and memory over time | Roles that improve with repetition |
| Gemini | Google's agent runtime | General |
| HTTP / webhook | Anything with an endpoint, including n8n workflows | Legacy automations, custom bots |
| Bash / process | Any local script | Glue jobs |
Dedicated guides: Paperclip + Claude Code, Paperclip + OpenClaw and Paperclip + Hermes Agent.
How to Install Paperclip AI
The fastest path is a single command on a machine with Node.js 24.11 or newer:
npx paperclipai onboard --yes
That downloads the app, starts an embedded PostgreSQL instance, walks you through creating your first company and opens the dashboard at http://localhost:3100. The project also publishes a checksummed install script:
curl -fsSLO https://paperclip.ing/install.sh
curl -fsSLO https://paperclip.ing/install.sh.sha256
shasum -a 256 -c install.sh.sha256
bash install.sh
To try it without installing anything permanent, use the isolated test drive with your own API key:
ANTHROPIC_API_KEY=sk-ant-... npx paperclipai test-drive
Production deployments usually run on a small VPS with Docker or Docker Compose, a real PostgreSQL volume and the dashboard behind Tailscale or an authenticated reverse proxy. Never expose port 3100 directly to the internet. Full instructions, Windows notes and the errors we hit most often are in How to install Paperclip AI and Paperclip AI hosting and self-hosting guide.
Integrations
Paperclip connects outward in three ways: adapters (the agents themselves), plugins (app integrations) and the API.
- Linear: bidirectional issue sync with comments and status mirroring, so your human team sees agent tickets in the tool they already use.
- Slack and Telegram: notifications when a ticket is created, completed or needs your approval.
- n8n: an n8n workflow can be hired as an agent through the HTTP adapter, or n8n can call Paperclip's API to create tickets from outside events. This is the combination we deploy most often; see our n8n services.
- OpenRouter: route OpenCode or Pi harnesses to any model provider.
- MCP: agents bring their own MCP tool servers, so Paperclip inherits whatever your agents can already reach.
Setup details for each are in Paperclip AI integrations: n8n, Slack, Telegram, Linear, OpenRouter and MCP.
Real Use Cases for Businesses
Most coverage of Paperclip focuses on the "zero-human company" idea. The practical wins we see are smaller and more valuable:
- Marketing agency content operations. A strategist agent turns a monthly plan into tickets, writer agents draft, an editor agent reviews against a SKILL.md style guide, and a human approves before anything publishes.
- E-commerce operations. Agents watch inventory feeds, draft supplier emails, reconcile orders and escalate exceptions to a person through Slack.
- Software shops. A CTO agent breaks features into tickets, engineer agents work in isolated git worktrees, a reviewer agent gates merges, and humans approve releases.
- Back-office for local businesses. Intake, follow-ups, reporting and data entry roles that previously lived in a dozen disconnected Zapier or Make scenarios get a single org chart, budget and audit log.
- Research and monitoring. Routines wake a research agent every morning to compile competitor, review or ad-performance summaries.
More examples, plus how company templates, skills, plugins and ClipMart fit in, are in Paperclip AI use cases, templates and ClipMart.
Pricing: Is Paperclip AI Free?
Yes. Paperclip is MIT-licensed, self-hosted, and does not require a Paperclip account. A hosted cloud version is on the roadmap but was not generally available as of mid-2026. Your real costs are:
- Hosting. A 2 vCPU / 4 GB VPS is enough for a small company of agents. Several hosts now sell one-click Paperclip templates.
- LLM usage. This is the variable. Per-agent budgets with auto-pause keep it bounded.
- Your time. Designing roles, writing skills and reviewing approvals is real work, especially in the first month.
We break down three sample monthly budgets in Paperclip AI pricing: the real cost of running an agent company.
Paperclip AI vs the "Paperclip Maximizer"
A large share of searches for "paperclip AI" are actually looking for the paperclip maximizer, a thought experiment philosopher Nick Bostrom described in 2003. In it, a superintelligent AI told to make paperclips converts everything, eventually including people, into paperclips, because nobody told it not to. It is a parable about goal misalignment and instrumental convergence, not a product.
The Paperclip app takes its name from that joke, and its features are arguably the practical answer to the worry: hard budgets, approval gates, pause and terminate controls and an immutable audit log are exactly the constraints the thought experiment says a runaway system lacks. If you came here for the philosophy, read The AI paperclip problem explained. Two other unrelated things also share the name: Universal Paperclips, a 2017 browser game by Frank Lantz, and Clippy, Microsoft's 1997 Office assistant.
Alternatives and Comparisons
Because Paperclip is a management layer, most "alternatives" are really neighbours. The honest comparison:
| Tool | What it is | Relationship to Paperclip |
|---|---|---|
| OpenClaw | Personal-assistant agent gateway with 25+ messaging channels and a big skills ecosystem | An employee Paperclip can hire. Best for solo operators on its own. |
| Hermes Agent | Nous Research agent with a self-improving skill and memory loop | An employee Paperclip can hire. Best when a role should get better with repetition. |
| Multica | Orchestrator launched April 2026 supporting eight agent runtimes | The closest direct competitor. |
| CrewAI, AutoGen, LangGraph | Python frameworks for building multi-agent systems in code | Frameworks, not management. You could run a CrewAI crew as one Paperclip agent. |
| Dify, MetaGPT | LLM app builder; role-based software-company simulation | Overlapping ideas, different audiences. |
| n8n | Visual workflow automation | Complement. Deterministic pipelines in n8n, judgment work in Paperclip. |
Full write-ups: Paperclip AI vs OpenClaw, 9 best Paperclip AI alternatives and our balanced Paperclip AI review 2026.
Should Your Business Use Paperclip?
Good fit if you already run two or more agents or automations that need to coordinate, if you want hard spending caps and an audit trail before you let agents touch customer data, or if you have a technical person (or an agency) who can own a self-hosted server.
Poor fit if you need one assistant, if nobody on your side is comfortable with a terminal and a VPS, or if your processes are fixed step-by-step sequences that n8n or Make already handle. Paperclip is a young project with a fast release cadence; independent reviews rate it around 8 out of 10, praising the dashboard and governance model while noting bugs and a learning curve.
How Neon Digital Media Deploys Paperclip for Clients
We have shipped business automation for Las Vegas law firms, medical practices, restaurants, home-services companies and e-commerce brands using n8n, Make, Zapier and custom Python. Paperclip slots into that stack as the layer that manages agents when a project outgrows single workflows. A typical engagement:
- Automation audit. We map your current workflows and decide which tasks are deterministic (keep in n8n) and which need judgment (assign to agents).
- Org design. We design the company, roles, budgets, approval gates and skills, and pick the right adapter for each role.
- Deployment. Hardened VPS, Docker, persistent Postgres, Tailscale or authenticated proxy, backups, secrets.
- Integration. Slack or Telegram approvals, Linear or your PM tool, CRM and e-commerce connections through custom API integration.
- Managed operation. We monitor spend, review audit logs, tune skills and handle updates each release.
Call +1 (702) 266-3865 or request a consultation to talk through whether an agent company makes sense for your business.
Go Deeper: Every Guide in This Series
Learn
- How Paperclip AI agent orchestration works: org charts, heartbeats, tickets and budgets
- Paperclip AI on GitHub: the paperclipai/paperclip repo explained
- The AI paperclip problem (paperclip maximizer) explained
- Paperclip AI use cases, company templates, skills, plugins and ClipMart
Set up
- How to install Paperclip AI: npx, Docker, Docker Compose and VPS
- Paperclip AI hosting and self-hosting guide
- Paperclip AI pricing: is it free, and what does it really cost?
Integrate
- Paperclip AI + OpenClaw integration guide
- Paperclip AI + Hermes Agent integration guide
- Paperclip AI + Claude Code: managed coding teams
- Paperclip AI integrations: n8n, Slack, Telegram, Linear, OpenRouter and MCP
Decide
Frequently Asked Questions
What is Paperclip AI?
Paperclip AI is an open-source, MIT-licensed platform that manages teams of AI agents like employees. It provides an org chart, goals, a ticket system, scheduled heartbeats, per-agent budgets, approval workflows and an audit log, and it works with agents such as Claude Code, Codex, Cursor, OpenClaw and Hermes.
Who created Paperclip AI?
It was created by a pseudonymous developer known as @dotta and is maintained by Paperclip Labs, Inc. It launched in early March 2026 after dotta grew frustrated running more than twenty uncoordinated Claude Code sessions for an automated trading operation.
Is Paperclip AI free?
Yes. It is free and open source under the MIT license and runs on your own machine or server. You pay only for hosting and for the LLM usage your agents generate. A hosted cloud edition is on the roadmap.
How do I install Paperclip AI?
On a machine with Node.js 24.11 or newer, run npx paperclipai onboard --yes. The dashboard opens at http://localhost:3100. There is also a checksummed install script at paperclip.ing/install.sh, and Docker deployment for servers.
What is the Paperclip AI default port?
The API and dashboard listen on port 3100 (http://localhost:3100) by default.
Which AI models does Paperclip support?
Paperclip is model-agnostic. It runs whatever your agents run: Claude through Claude Code, OpenAI models through Codex, Gemini, any OpenRouter model through OpenCode or Pi, and any custom HTTP agent.
What is the difference between Paperclip AI and OpenClaw?
OpenClaw is an agent: a personal-assistant gateway that talks over WhatsApp, Telegram, Slack and Discord and has a large skills ecosystem. Paperclip is the company layer that hires, schedules, budgets and audits agents like OpenClaw. Many teams run both.
Is Paperclip AI the same as the paperclip maximizer?
No. The paperclip maximizer is a 2003 thought experiment by Nick Bostrom about a misaligned superintelligence. The Paperclip app borrows the name as a joke, and its budgets, approvals and audit logs are practical safeguards against exactly that kind of runaway behaviour.
Is Paperclip AI safe to run?
It is as safe as your deployment. Keep the dashboard off the public internet (Tailscale or an authenticated reverse proxy), store keys in Paperclip's encrypted secrets, set conservative per-agent budgets, and require approvals for anything that touches production systems or customer data.
What is ClipMart?
ClipMart is the official marketplace for portable Paperclip company templates, still rolling out as of September 2026. A separate third-party site, ClipMarts.com, sells community skills and templates and is not affiliated with Paperclip Labs.
Does Paperclip AI work with n8n?
Yes, in both directions. An n8n workflow can be hired as an agent through the HTTP adapter, and n8n can create or update Paperclip tickets through the API. Neon Digital Media builds and manages this stack for businesses.
Can Neon Digital Media set up Paperclip for my business?
Yes. We audit your workflows, design the agent org chart and budgets, deploy a hardened self-hosted instance, integrate it with your tools and manage it month to month. Call +1 (702) 266-3865 to get started.