Installation guide

Paperclip setup checklist.

Paperclip is a free, open-source tool that lets you manage a team of AI agents, like a staff of assistants that research, write and carry out tasks on autopilot. This checklist walks you through every step from installation to your first working agent. No coding experience needed.

7Steps
110k+GitHub stars
0€Licence cost
1Command to install

APrerequisites

A1

Open a terminal

Prereq
Why you need it
The terminal is where you type commands to install and run Paperclip. It looks like a text-only screen: no buttons, just typing. Sounds scary, but you only need it for a few minutes.
What to do
Mac: open Spotlight (Cmd+Space), type "Terminal", press Enter. Windows: press Win+R, type "cmd", press Enter. Or use the built-in terminal in VS Code (Terminal > New Terminal).
A2

Install Node.js

Prereq
Why you need it
Paperclip runs on Node.js, a free program that lets your computer run modern tools. Without it, the install command won't work.
What to do
Go to nodejs.org, download the current LTS version and run the installer. To confirm it worked, type this in your terminal:
node --version
You need v24.11 or higher. An older version will make the install fail.
A3

Install Claude Code and sign in

Prereq
Why you need it
Paperclip doesn't run the AI itself. Its default agent type, "Claude Code", hands each task to the Claude Code app on your computer. Paperclip reuses that sign-in, so you don't have to set it up twice.
What to do
Install Claude Code by following the steps at claude.com/claude-code. On a Mac you can paste:
curl -fsSL https://claude.ai/install.sh | bash
Then type claude and sign in with your Claude subscription. No subscription? Create an API key in the Claude Console instead and pay per use. Costs per use

BOne command

B

Install Paperclip

Install
Why you need it
This single command downloads the latest Paperclip and sets up everything on your computer, including its own database. It takes a few minutes.
What to do
Paste this into your terminal and press Enter:
npx paperclipai@latest onboard --yes
Wait for it to finish. To start Paperclip again later, run npx paperclipai run.

CYour control panel

C

Open the dashboard

Install
Why you need it
The dashboard is Paperclip's visual control panel. It opens in your browser like a website, but runs locally on your machine. This is where you'll manage everything from now on.
What to do
If it doesn't open by itself, go to:
http://localhost:3100
Follow the setup steps on screen.

DCreate your company

D1

Name your team

Config
Why you need it
Paperclip calls your team a "company", and every agent in it is an employee. Think of it as a restaurant: the company is the restaurant, the agents are the kitchen staff. The name helps you stay organised if you create more teams later.
What to do
Click New Company and type any name that makes sense to you, plus a short description. Examples: "Content Team", "Marketing Ops" or "Lu's AI Kitchen". You can change it later.
D2

Set a goal

Config
Why you need it
The goal tells your agents what the team exists to do, and every task traces back to it. Vague goals produce vague work. Specific goals produce specific results.
What to do
Write one clear sentence. Good: "Produce one LinkedIn post per week about AI tools for consultants." Bad: "Do content stuff." The more specific, the better your agents perform.

EHire the head chef

E1

Create your first agent

Config
Why you need it
Every team needs a leader. Your first agent is the "head chef" (Paperclip calls it the CEO). It reads your goal, breaks it into tasks and delegates work to agents it hires later.
What to do
Create the agent with the role CEO and choose Claude Code as its adapter (how the agent runs). Name it "CEO" or "Head Chef". This is the only agent you have to create yourself.
E2

Pick the AI model

Config
Why you need it
The model is the brain behind your agent. Smarter models cost more per task, and Paperclip picks the top-tier Opus model unless you choose otherwise.
What to do
For a first team, choose the current Sonnet model in the model setting (Claude Sonnet 5.5 in November 2026). Best value It costs less than Opus and handles coordination well. You can switch any agent to Opus later.
E3

Test the connection

Config
Why you need it
This confirms your agent can actually reach Claude. If this fails, nothing else will work, so it's worth checking now.
What to do
Click Test Environment on the agent. You should see a success message. If it fails, check that Claude Code is installed and signed in (step A3), or that your API key is correct.
E4

Give it a task

Config
Why you need it
This is your first real test. You'll see the agent think, plan and produce output: proof that the whole system works.
What to do
Create a simple task, like "Write a 3-sentence summary of what our team does", and assign it to your CEO. Watch the agent work in real time. Your first AI employee is live.

FSafety controls

F1

Set a budget

Config
Why you need it
Without a budget, agents can burn through credits with no limit. A budget cap means you never wake up to a surprise bill.
What to do
Set a monthly budget for the company and for each agent. Start low, for example 20€, and raise it once you see what your agents actually use.
F2

Set a heartbeat

Config
Why you need it
Agents don't run all the time. They wake up in short "heartbeats": on a schedule, when they get a task or when someone mentions them. The schedule decides how often they check in on their own. Think of it as office hours for your AI staff.
What to do
In the agent's heartbeat settings, set the interval to 8 hours. Your agents wake up, do their tasks, then go back to sleep. Three work sessions a day, like a team working in shifts.
F3

Know the safety net

Config
Why you need it
This is the most important safety feature. At 80% of its budget an agent is told to stick to critical work. At 100% it stops completely. No overruns.
What to do
No action needed: this is on by default. When the budget runs out, the agent pauses automatically. You review what happened, then raise the budget to resume it, or it resumes when the new month starts.

GManage the team

G1

Approve the plan

Config
Why you need it
Your head chef doesn't just execute blindly. It proposes a strategy first and waits for your approval before work starts. You stay the decision-maker.
What to do
Open the Approvals page in your dashboard. You'll see your CEO's plan. Read it, then click Approve, Reject, or ask for a revision with notes.
G2

Approve new hires

Config
Why you need it
As tasks get more complex, your head chef will want specialist agents: a researcher, a writer, an editor. With hire approvals on, you approve or reject each one, like a real manager.
What to do
When a hire request appears in Approvals, review the proposed agent's name, role, model and budget. Approve it to add it to the team or reject it if it doesn't make sense. Start small: 3 to 5 agents is plenty.
G3

Watch the org chart

Config
Why you need it
The org chart shows your whole AI team at a glance: who reports to whom, who's working, who's idle. It updates automatically as agents are hired and tasks move through the system.
What to do
Open the org chart. You'll see your head chef at the top with any hired agents below. You're the restaurant owner now: the kitchen runs itself, and you watch from the pass.
Before you start

Start with one task.

Agent teams are an advanced setup. First define the work, the result you need and where a person must review it. Check the project's current documentation for installation requirements and model costs.

Have it set up for you → Read the introduction