Run a Small AI Team for Under €100 a Month With Paperclip
In short
- Paperclip is a free, open-source (MIT) app that runs a team of AI agents like a company: org chart, goals, tasks, budgets, approvals.
- Version v2026.1023.0 landed on October 23 with 200 commits: private tasks, your own model gateway (OpenRouter, Bedrock, Gemini), an exact cost ledger and self-installing Slack bots.
- The software costs €0. The agents don’t – Paperclip’s own docs say plan on $20–100 a month in model costs for an active team.
You have five AI tools open. Nobody knows what the others are doing. One writes posts. One does research. One answers email. And you are the glue – copying, pasting, checking. That’s not automation. That’s a second job.
The old fix? Hire people. A junior marketer, a VA, a researcher. The new fix is called Paperclip. You give it a goal, it gives you a team of AI agents with a boss, a task board and a budget. And the app itself costs nothing.
What Paperclip is
Paperclip is built by Paperclip Labs. It’s a Node.js server with a React dashboard that orchestrates AI agents. Their own one-liner says it best: if an agent like OpenClaw is an employee, Paperclip is the company.
It doesn’t build agents. It manages them. You plug in agents you already use – Claude Code, Codex, Cursor, Gemini CLI, OpenCode, OpenClaw, Kimi Code, Grok Build and more – and Paperclip gives them roles, reporting lines, tasks and spending limits. It looks like a task manager. Underneath it’s an org chart with governance.
The repo was created in March 2026. Today it has over 107,000 GitHub stars and more than 2,700 merged pull requests. That’s not a weekend project anymore.
What’s new in v2026.1023.0
- Private tasks and projects. Share work only with named people and agents. Others can’t see it in lists, search or run transcripts.
- Bring your own model gateway. Route agents through OpenRouter, Amazon Bedrock, Google Gemini or any OpenAI- or Anthropic-compatible endpoint. Cheaper models for cheap tasks.
- Costs as an exact ledger. Every charge is marked reported, estimated or unpriced. Your ChatGPT or Claude subscription fee shows up next to API spend.
- Budgets block unpriced usage by default. If Paperclip can’t price a run, the budget counts as maxed out until you resolve it. Safe by default.
- Slack bots that set themselves up through a four-screen wizard, plus 16 new apps in the catalog (Calendly, Firecrawl, Tavily, Resend, Hugging Face and others).
What it costs
How much is Paperclip? Nothing. It’s MIT-licensed, self-hosted, and you don’t even need a Paperclip account. A hosted Paperclip Cloud exists only as a waitlist – no public price yet.
The real bill is the AI behind the agents. Paperclip’s own quickstart is honest about it:
| Item | Cost |
|---|---|
| Paperclip (self-hosted) | Free, MIT licence |
| Paperclip Cloud | Waitlist, no price published |
| Testing it out (model costs) | $5–20 one-off, per Paperclip docs |
| An active AI company (model costs) | $20–100/month, per Paperclip docs |
Do the math: say you pay a freelancer for social posts, blog drafts and competitor research – a modest €600 a month (our example, not a market rate). That’s €7,200 a year. A Paperclip team at the top of their own range, $100 a month (under €100), costs less than €1,200 a year. Add a €10/month server and you’re at about €1,300. You keep roughly €5,900 a year. You still review the output – but you review, you don’t produce.
Step-by-step: your first AI team
- Get the basics ready. You need a computer or server with Node.js 24.11 or newer, plus an API key from Anthropic or OpenAI. For Claude-based agents, install Claude Code on the same machine.
- Install with one command. In a terminal run
npx paperclipai@latest onboard --yes. It starts the dashboard athttp://localhost:3100and creates its own database. No database setup on your side. - Lock it down if it’s not just your laptop. The quick install runs in trusted local mode. For access from other devices, use
--bind lanor--bind tailnetfor authenticated mode, or the official Docker image. - Create a company and write the goal. One clear goal, with a number. “Publish 3 SEO articles a week and grow organic traffic 20% in 90 days” beats “do marketing”.
- Set budgets before anything runs. Company budget, agent budget, project budget. Paperclip pauses agents automatically at 100%. Do this first – not after the first surprise invoice.
- Hire the CEO agent. It’s the first agent you add. It reads your goal and proposes a strategy.
- Approve the strategy. It lands in the Approvals queue. Read it. Edit it. Only then do tasks appear on the board.
- Add specialists. The CEO can request new hires through approvals: a writer, a researcher, a support agent. Use a strong model for the CEO and cheaper models via OpenRouter for routine work.
- Connect your tools. GitHub, Notion, Airtable, Slack, or your own MCP server. Give each agent only the connections it needs.
- Turn on heartbeats and routines. Agents wake for assigned tasks, follow-ups or schedules. A routine like “every Monday: competitor price check” runs on its own.
Paperclip says the path from a running install to a working agent takes about 5–11 minutes. Plan an afternoon to get the team right.
Real use cases for a small business
- Content team. Researcher finds topics, writer drafts, editor agent checks – you approve before anything goes live.
- E-commerce ops. For a parts shop like GiantParts: an agent drafts product descriptions from part numbers, another checks listings for missing data, a routine flags stock issues every morning.
- Support triage in Slack. A Slack agent picks up questions, creates tasks and reports back in the same thread.
- Code review. The GitHub review bot posts a “Paperclip Review” check with a pass score you set.
Where it breaks
- It’s a tool for builders. Terminal, Node.js, API keys. If you’ve never opened a terminal, expect a rough first hour.
- Model bills are on you. Ten agents with heartbeats on can burn money fast. Budgets help – but only if you set them.
- “Autonomous company” is marketing. Agents still make mistakes. The approval queue is not optional; it’s the product.
- Fast release pace. 200 commits in one release, with breaking changes – this time the Operator role went from 1 permission to 14. Read release notes before you update.
- Self-hosting is your job. Backups, security, uptime. Cloud is still a waitlist.
- Some features are experimental. The GitHub review bots now sit behind an experimental switch, as do new Runner providers like OpenAI Dot and Pi 1.0.
Alternatives
- n8n – visual workflows with AI agent nodes; free to self-host. Better for fixed processes, weaker at “team with a goal”.
- CrewAI – open-source Python framework for multi-agent crews. More control, but you write code.
- Make – no-code scenarios with AI modules. Easiest start, but you design every step yourself.
Ivan’s Take
This is the right idea: stop prompting tools one by one and start managing a team with goals and budgets. It’s free, it’s open and it’s moving fast – worth an afternoon if you already use Claude Code or Codex. I’d start with one small team, a hard €50 budget and every output going through approval, and only scale what actually saves hours.
Sources: Paperclip (GitHub) · Paperclip
If it doesn’t save you time or money, you won’t see it here
- Facts, not hypeEvery price and number is checked against the official source
- Always the real priceYou see what it costs before you click
- I say when it’s weakEven when it’s OpenAI or Google
