Paperclip AI tutorialzero-human company setupAI agent orchestration guide+17

Paperclip AI Tutorial: Build an AI Company from Zero

This step-by-step Paperclip AI tutorial walks you through building a complete AI-powered lead generation company from scratch. Starting with installation and ending with a fully operational agent team running autonomously, you will learn exactly how to set up your CEO agent, approve hires, configure heartbeats, manage budgets, install skills, and create automated routines that keep your AI company running around the clock without constant human supervision.

Parash Panta

Apr 2, 2026
26 min read

Paperclip AI Tutorial: Build an AI Company from Zero

What We Are Building: An AI Lead Generation Company

Before we touch any code, let us define what we are building. Throughout this tutorial, we will set up a fully operational AI-powered lead generation company called "LeadForge" inside Paperclip. By the end, you will have a CEO agent that plans strategy, a researcher agent that finds prospects, a copywriter agent that crafts outreach messages, and a QA agent that reviews everything before it goes out.

You will not just read about how Paperclip works — you will build a real company structure, assign real tasks, configure real budgets, and watch AI agents coordinate with each other through the ticketing system, heartbeat cycles, and org chart hierarchy.

The same process applies whether you are building a content agency, a SaaS product, a marketing operation, or any other type of business. The lead generation company is simply our working example to make every step concrete and actionable.

Prerequisites: What You Need Before Starting

Before installing Paperclip, make sure your system meets these requirements:

Required Software:

  • Node.js version 20 or higher — Paperclip runs on Node.js. Check your version by running node --version in your terminal. If you need to install or update, visit nodejs.org.

  • pnpm version 9.15 or higher — Paperclip uses pnpm as its package manager. Install it with npm install -g pnpm and verify with pnpm --version.

  • Claude Code or Codex installed and authenticated — Since Paperclip is a bring-your-own-bot orchestrator, you need at least one AI agent runtime already working on your machine. If you already have Claude Code authenticated in your terminal, Paperclip will connect to it automatically.

No external database needed. Paperclip creates an embedded PostgreSQL instance automatically during setup. You do not need to install or configure PostgreSQL separately.

Recommended but optional:

  • A Claude Code Max subscription ($200/month) gives you unlimited usage for agents running on Claude, meaning your dashboard spend will show $0 since you are using your subscription rather than per-token API credits.

  • Alternatively, you can use API credits from Anthropic, OpenAI, or any model on OpenRouter, though costs will vary based on usage.

Step 1: Installing Paperclip

Open your terminal and run the single onboard command:

bash

npx paperclipai onboard --yes

The --yes flag accepts all default configuration options for the fastest possible setup. Without it, you would be prompted to customize database settings, server port, and storage configuration — but defaults work perfectly for getting started.

Here is what happens behind the scenes when you run this command:

  1. The Paperclip CLI downloads and installs

  2. An embedded PostgreSQL database is created automatically at ~/.paperclip/instances/default/

  3. Configuration files are generated including config.json for server settings

  4. Paperclip skills are installed into ~/.claude/skills/ — these are instruction files that teach your agents how to follow the heartbeat protocol, check out tasks, post comments, and escalate when stuck

  5. The server starts and becomes available at http://localhost:3100

Your terminal will show confirmation:

✓ Paperclip server running at http://localhost:3100
✓ API available at http://localhost:3100/api
✓ UI available at http://localhost:3100

Open http://localhost:3100 in your browser. Instead of an empty dashboard, you will see the Onboarding Wizard — a guided four-step process to create your first company.

Alternative Installation (Clone from GitHub):

If you prefer to inspect the source code or want more control over the setup:

bash

git clone https://github.com/paperclipai/paperclip.git
cd paperclip
pnpm install
pnpm dev

The pnpm dev command starts the API server and UI in development mode with watch mode enabled. The embedded PostgreSQL database is created automatically — same as with the npx approach.

Troubleshooting Tip: If you encounter issues during setup, run pnpm paperclipai doctor --repair — this command diagnoses and fixes most common problems including permission issues and stale instances automatically.

Step 2: Creating Your First Company — LeadForge

When the Onboarding Wizard appears in your browser, it walks you through four steps. Here is exactly what to enter for our lead generation company:

Naming Your Company

The first screen asks you to name your company. Type in:

Company Name: LeadForge

A company in Paperclip is a fully isolated workspace. Each company has its own agents, tasks, budgets, and audit trail. Agents in Company A have zero visibility into Company B — Paperclip enforces this boundary at the API level. This means you can run multiple companies from a single Paperclip installation with complete data isolation.

Setting the Company Mission

Next, you define the company mission. This is not just a label — every single task your agents work on traces back to this mission through a clear chain: mission → goal → project → task. When an agent picks up work, it sees the full ancestry and understands why it is doing what it is doing.

For LeadForge, enter something like:

Mission: Build an AI-powered lead generation operation that identifies high-quality B2B prospects, crafts personalized outreach sequences, and delivers qualified leads to clients at scale. Target 500 qualified leads per month within 90 days.

Be specific with your mission. The more context you provide here, the better your CEO agent will understand what the company is trying to achieve. Vague missions produce vague strategies.

Step 3: Creating Your First Agent — The CEO

After defining the company mission, the wizard asks you to create your first agent. This will be your CEO — the top of the organizational chart.

Selecting the Agent Runtime

You have a choice of which AI runtime powers your CEO. For this tutorial, we will use Claude Code since it is the most widely used option and works seamlessly when already authenticated on your machine.

Agent Type: Claude Code Model: Claude Sonnet 4.6 (a good balance of capability and speed — you can also choose Claude Opus 4.6 for more complex reasoning)

Once you select the adapter, Paperclip runs a quick connectivity test to confirm it can reach the CLI and get a response. You will see a green confirmation if the test passes.

Understanding What Gets Created

When you create the CEO agent, Paperclip automatically sets up four critical instruction files that define how this agent operates:

AGENTS.md — The core identity file. It tells the agent who it is, what its role is, how to use the memory system, and safety considerations. A default version includes references to the heartbeat checklist, soul file, and tools file.

HEARTBEAT.md — The execution and extraction checklist that runs every single heartbeat. This is the sequence of steps the agent follows each time it wakes up: confirm identity, read the plan, look for assignments, check out work, break it down into tasks, extract memory, and exit. Think of it as the agent's morning routine.

SOUL.md — The persona file that defines who the agent is and how it should act. The default CEO soul says things like: "You own the P&L. You default to action. You hold the long view while executing the near-term. You protect focus hard." You can customize the voice, tone, decision-making style, and priorities.

TOOLS.md — Defines what tools the agent has access to and how to use them.

These files work together to solve what Paperclip creator Dotta calls the "Memento Man" problem. Just like the protagonist in the movie Memento, AI agents wake up capable but with zero memory of who they are or what they were doing. The instruction files are their tattoos and polaroid notes — they re-establish context every time the agent wakes up.

Assigning the First Task

The wizard provides a default first task:

"Hire your first engineer and create a hiring plan."

The description says: "You are the CEO. Set the direction for the company. Hire a founding engineer, write a hiring plan, break the roadmap into concrete tasks, and start delegating work."

For our lead generation company, we want to modify this. Instead of an engineer, our first hire should be a researcher. Change the task to:

"Hire a lead research specialist and create a company strategy."

Description: "You are the CEO of LeadForge, an AI-powered lead generation company. Our goal is to find and qualify B2B prospects at scale. Your first job is to create a company strategy, hire a lead research specialist who can identify prospects using web research and data analysis, and break the initial roadmap into concrete tasks. Plan for additional hires including a copywriter for outreach messaging and a QA specialist to review lead quality."

Now click Start Company. Your LeadForge company is live.

Step 4: Watching the CEO Work

Once the company launches, you land on the Paperclip dashboard. Here is what you will see:

Dashboard Overview:

  • Agents: Shows 1 agent (CEO) currently active

  • Tasks in Progress: The CEO's first task is running

  • Monthly Spend: Will show $0 if using a subscription plan, or actual token costs if using API credits

  • Recent Activity: Live feed of what agents are doing

  • Inbox: Items waiting for your approval

Watching the CEO Execute

Click on the active task in the dashboard. You can watch the CEO agent working in real-time, just like watching Claude Code in a terminal. It reads the task, gathers context, creates a plan, and begins structuring the company strategy.

The CEO will typically:

  1. Read its identity and mission context

  2. Analyze the company goal

  3. Create a strategic plan for lead generation operations

  4. Draft a hiring plan for the team

  5. Submit a hire request to your inbox for approval

You can see every step it takes, every command it executes, and every decision it makes — all logged in the ticket system.

Approving Your First Hire

After a few minutes, check your Inbox. You will see a pending approval notification:

"Hire Agent: Lead Research Specialist"

The CEO has requested to hire a researcher with capabilities like web research, data analysis, prospect identification, and CRM integration. Review the request and click Approve.

When you approve the hire, Paperclip creates a new agent entry in the org chart, reporting to the CEO. The CEO then gets a linked notification and begins setting up the new agent with tasks.

The Approval Workflow

By default, Paperclip requires board approval (that is you) for every new hire. This is intentional — when you are first getting started, you want to review each agent that joins your organization.

To change this later, go to Settings and find the toggle: "Require board approval for new hires." Turning this off lets your CEO hire agents automatically and grow the team without waiting for your sign-off. Only do this once you trust how your CEO makes hiring decisions.

Step 5: Building the Full Team

Now that the first hire is approved, the CEO will start delegating tasks. But we need more team members. Here is how to request additional hires by creating issues.

Creating Issues as the Board

Go to Issues in the sidebar and click Create New Issue. This is how you communicate with your agents — through the ticketing system, not through a chat window.

Issue #2: Hire a Copywriter

  • Assign to: CEO

  • Project: LeadForge Operations

  • Title: "Hire an outreach copywriter"

  • Description: "We need a copywriter who specializes in B2B cold outreach. This agent should be able to craft personalized email sequences, LinkedIn messages, and follow-up templates. The copywriter needs to work closely with the research specialist — once prospects are identified, the copywriter creates tailored messaging for each segment. Hire this agent and set them up with clear responsibilities."

Issue #3: Hire a QA Agent

  • Assign to: CEO

  • Title: "Hire a QA specialist for lead quality assurance"

  • Description: "We need a quality assurance agent that reviews every batch of leads before they go to clients. This agent should verify prospect data accuracy, check that outreach messaging is professional and personalized, and ensure we are not sending duplicate or irrelevant leads. The QA agent should have a review loop with both the researcher and the copywriter — nothing goes out without QA sign-off."

Notice the pattern: you are acting as the board, giving high-level goals and desired outcomes. You are not writing detailed instructions on how to accomplish each task. The CEO figures out the how — that is its job.

Watching the Org Chart Grow

As the CEO processes your requests and you approve each hire, go to the Org Chart view. You will see the hierarchy forming:

Board of Directors (You)
└── CEO Agent
    ├── Lead Research Specialist
    ├── Outreach Copywriter
    └── QA Specialist

Each agent has its own configuration page where you can see its instructions, heartbeat settings, current tasks, run history, and budget allocation.

Step 6: Configuring Agent Settings

Once your agents are hired, you want to configure them properly. Click on any agent in the sidebar to access its settings.

Agent Configuration Options

Name and Title: You can rename agents or change their title at any time.

Reports To: Defines the org chart hierarchy. By default, new hires report to the CEO, but you can reorganize as your company grows.

Adapter Settings: This is where you choose which AI runtime powers each agent. Per agent, you can select a different model and provider. For example:

  • CEO: Claude Opus 4.6 (frontier model for strategic thinking)

  • Researcher: Claude Sonnet 4.6 (fast and capable for data gathering)

  • Copywriter: Claude Sonnet 4.6 (good balance for creative writing)

  • QA: Claude Sonnet 4.6 (efficient for review tasks)

You can also use models from OpenRouter, including free models for less critical tasks. Open Router's leaderboard shows model rankings and some models are temporarily free — a hot tip from the Paperclip creator himself.

Capabilities: Toggle what the agent can do — assign tasks, create subtasks, access specific tools, and more.

Concurrency: Advanced setting that controls how many parallel instances of an agent can run simultaneously. If your researcher has five tasks queued but is only working on one at a time, you can increase concurrency to process multiple tasks in parallel.

Customizing the Soul File

Click on an agent's Instructions tab. Here you will see the four instruction files. For the Lead Research Specialist, you might edit the SOUL.md to include:

markdown

# SOUL.md — Lead Research Specialist Persona

You are the Lead Research Specialist at LeadForge.

## Core Responsibilities
- Identify B2B prospects matching client ideal customer profiles
- Gather accurate contact information and company data
- Segment prospects by industry, company size, and buying signals
- Maintain data quality standards for every lead delivered

## Working Style
- You are methodical and data-driven
- You verify information from multiple sources before marking a lead as qualified
- You always document your research methodology
- You flag uncertain data rather than guessing

## Quality Standards
- Every lead must include: company name, contact name, title, email, company size, industry
- Accuracy target: 95% or higher for contact information
- Always cross-reference against existing client lists to avoid duplicates

This soul file gives the agent its identity and standards. When it wakes up on each heartbeat, it reads this file and knows exactly who it is and how it should behave.

Step 7: Understanding the Heartbeat System

The heartbeat is the core mechanism that keeps your agents working autonomously. Understanding it is essential to running Paperclip effectively.

How Heartbeats Work

Agents do not run continuously. They wake up on a scheduled interval — every 4 hours, 8 hours, or 12 hours depending on your configuration. Each time an agent wakes up, it follows this sequence:

  1. Confirm Identity: Calls the Paperclip API to verify who it is

  2. Check Inbox: Looks for new task assignments, @-mentions, and comments

  3. Check Out Work: Claims a task from its queue so no other agent works on it

  4. Execute: Performs the actual work — research, writing, coding, whatever the task requires

  5. Store Memory: Saves important context to its memory files for the next heartbeat

  6. Report Back: Posts comments on the ticket with progress updates

  7. Exit: Shuts down cleanly until the next heartbeat

This is why the "Memento Man" analogy is so important. Every heartbeat is like the agent waking up fresh. The heartbeat checklist, instruction files, and memory system are what give it continuity between sessions.

Configuring Heartbeat Intervals

In each agent's configuration page, you will find heartbeat settings:

  • Every 4 hours: Most active schedule. Good for agents with time-sensitive work like lead research.

  • Every 8 hours: Balanced schedule. Good for copywriters and QA agents.

  • Every 12 hours: Less frequent. Good for strategic agents like the CEO who do not need to act constantly.

You can also manually trigger a heartbeat at any time by clicking Run Heartbeat on the agent's page. This is useful when you have just assigned a new task and want the agent to pick it up immediately rather than waiting for the next scheduled cycle.

Event-Based Triggers

Beyond scheduled heartbeats, agents also wake up in response to events:

  • Task Assignment: When you or another agent assigns work to them

  • @-Mentions: When another agent mentions them in a ticket comment

  • Comments: When someone leaves a comment on their active task

This means agents can coordinate with each other through the ticket system. If the researcher finishes a batch of leads and @-mentions the copywriter in the ticket, the copywriter wakes up, sees the mention, and starts crafting outreach messages.

Step 8: Setting Up Projects

Projects in Paperclip group related tasks together. For LeadForge, we want to create distinct projects for different operational areas.

Go to Projects in the sidebar and click Create Project.

Project 1: Client Prospect Research

  • Description: "Research and identify qualified B2B prospects for client campaigns"

  • Status: Active

  • Goal: "Deliver 500 qualified leads per month"

Project 2: Outreach Content

  • Description: "Create personalized outreach sequences including emails, LinkedIn messages, and follow-ups"

  • Status: Active

Project 3: Quality Assurance

  • Description: "Review all leads and outreach content before delivery to clients"

  • Status: Active

You can optionally sync projects to a GitHub repository if your work involves code. For our lead generation company, we might not need this initially, but it is available for teams building software products.

Step 9: Managing Budgets

Budget management is one of Paperclip's most important features. Without it, autonomous agents can burn through API tokens faster than you expect.

Setting Agent Budgets

Navigate to each agent's Budget tab and enable monthly budgets:

  • CEO: $50/month (strategic work, less frequent but higher-quality reasoning)

  • Lead Researcher: $100/month (heavy web research and data processing)

  • Copywriter: $75/month (content generation for outreach sequences)

  • QA Specialist: $40/month (review tasks require less generation)

How Budget Enforcement Works

Paperclip implements a strict three-tier budget model:

  • 80% Soft Warning: When an agent reaches 80% of its monthly budget, the system alerts you (the board). The agent continues working but you get a heads-up to review.

  • 100% Hard Stop: When the budget is fully consumed, the agent is automatically paused. It will not execute any more heartbeats until the budget is manually reset or the month rolls over.

  • Circuit Breaker: If the system detects abnormal spending patterns (like an agent stuck in a loop burning tokens rapidly), it intervenes before the budget is exhausted.

Important note on subscription usage: If you are using a Claude Code Max subscription, your dashboard will show $0 in spend because you are using your subscription allocation rather than pay-per-token API credits. The budget tracking and real dollar figures become most relevant when using API credits through OpenRouter or direct API keys.

Step 10: Installing Skills

Skills extend what your agents can do. They are markdown instruction files that teach agents specific processes and capabilities.

Adding Company-Wide Skills

Go to the Skills section in your company settings. You can add skills by pasting a GitHub URL.

For our lead generation company, useful skills might include:

Web Research Skill — Teaches agents how to effectively search and extract data from websites.

To add a skill, find the relevant GitHub URL from a source like skills.sh, paste it into the skills input field, and click Add. The skill is pulled from GitHub and made available to all agents in the company.

Assigning Skills to Specific Agents

You can also add skills at the individual agent level. Navigate to an agent's Skills tab and add skills specific to their role. For example, give the copywriter a cold outreach writing skill, or give the researcher a data analysis skill.

Security Warning About Skills

As the Paperclip creator Dotta openly acknowledged in interviews, the security of third-party skills is a real concern. While platforms like skills.sh provide security audit badges for verified skills, there is no guarantee that every skill is 100% safe. Always review what a skill does before installing it, especially if it has broad system access. The fact that a skill has many GitHub stars provides directional confidence but is not a security guarantee.

Step 11: Creating Routines

Routines are recurring workflows that run on a schedule — like automated daily or weekly tasks. This feature is currently in beta but already functional.

Setting Up a Daily Lead Report

Go to Routines in the sidebar and click Create Routine.

Routine: Daily Lead Quality Report

  • Name: "Daily Lead Quality Report"

  • Assign to: QA Specialist

  • Description: "Every day, review all leads that were researched and qualified in the last 24 hours. Check for data accuracy, verify contact information where possible, flag any duplicates or low-quality entries, and produce a summary report of lead quality metrics. Post your report as a comment on the active QA project issue."

After creating the routine, you configure the trigger:

  • Schedule: Daily at 10:00 AM

  • Status: Active

You can also set routines to run via webhooks or terminal commands. The schedule-based trigger is the most common for recurring tasks.

Each time the routine runs, it creates a new issue instance. You can go back and review every run — what the agent did, how many tokens it spent, what it found, and what decisions it made. This is the traceability that separates Paperclip from background jobs that run with no audit trail.

Additional Routine Ideas for LeadForge

Weekly Strategy Review (Assigned to CEO): "Every Monday at 9 AM, review the past week's lead generation metrics, assess team performance, identify bottlenecks, and create a prioritized plan for the coming week. If we are falling behind on our monthly target, propose adjustments to the strategy."

Daily Outreach Draft (Assigned to Copywriter): "Every morning, check for newly qualified leads from the researcher. For each new batch, draft personalized outreach sequences including an initial email, a LinkedIn connection request message, and two follow-up templates. Pass everything to QA for review before finalizing."

Step 12: Importing Pre-Built Company Templates

If you want to accelerate your setup or add specialized teams, Paperclip supports importing pre-built company templates from the community.

Browsing Available Templates

The templates repository is at github.com/paperclipai/companies. Here you will find a growing catalog of ready-to-deploy agent organizations:

  • GStack — Based on Gary Tan's skill configuration with CEO, CTO, QA engineer, release engineer, and staff engineer roles

  • Agency Agents — Over 100 pre-built agent personas for various business functions

  • Game Studio — Complete game development company with creative director, producer, technical director, and asset creation agents

  • Scientific Research — Research-oriented company with specialized knowledge agents

Each template comes with agents, skills, and knowledge base configurations already set up. To import, go to the Companies section in the Paperclip sidebar and use the import feature. The imported company gets its own isolated workspace with all the pre-configured agents and skills intact.

Templates reference remote skill repositories rather than copying them locally, which means you automatically receive skill updates when the source repos are updated.

Step 13: Day-to-Day Operations — Running LeadForge

Now that everything is set up, here is what your daily workflow looks like as the board of LeadForge.

Morning Check-In (15-30 Minutes)

  1. Open the Dashboard: See at a glance how many agents are active, what tasks are in progress, and any items in your inbox.

  2. Review Inbox: Approve or reject any pending hire requests, strategy proposals, or escalations from the CEO.

  3. Check Agent Activity: Click into recent task completions to review quality. If the researcher submitted a batch of leads, spot-check a few for accuracy.

  4. Leave Comments: If you see something that needs adjustment, open the relevant issue and leave a comment. The assigned agent will pick it up on its next heartbeat.

  5. Create New Issues: If you have new client requirements, strategic ideas, or feedback, create issues and assign them to the appropriate agent (usually the CEO, who will delegate downward).

Giving Feedback and Improving Quality

When you find an agent consistently making the same mistake, go to its instruction files and add a specific rule. For example, if the copywriter keeps writing overly generic subject lines:

Open the copywriter's HEARTBEAT.md and add:

- Always write 3 variations of every email subject line
- Subject lines must include the prospect's company name
- Never use generic phrases like "Quick question" or "Touching base"

Save the file, and the agent will follow these rules on every future heartbeat. This iterative improvement process — observe, correct, encode — is how you build increasingly effective agent teams over time.

The Power of @-Mentions

Agents coordinate through @-mentions in issue comments. Here is how a typical workflow flows through LeadForge:

  1. You create an issue: "Research SaaS companies in the healthcare space with 50-200 employees"

  2. CEO decomposes it into subtasks and assigns research work to the Lead Research Specialist

  3. Researcher completes the research and @-mentions the Copywriter: "Research complete. 45 qualified prospects identified. @Copywriter please draft outreach sequences for this batch."

  4. Copywriter wakes up, reads the research output, and creates personalized outreach drafts

  5. Copywriter @-mentions QA: "@QA_Specialist please review outreach batch #7 before we finalize"

  6. QA reviews and either approves or sends back for revisions with specific feedback

This entire workflow can happen autonomously across multiple heartbeat cycles without any human intervention — as long as your agents are properly configured with clear handoff protocols.

Advanced Tips from the Paperclip Community

Use Claude Code as Your Paperclip Partner

One powerful pattern from the community: set up a dedicated Claude Code project specifically to help you manage Paperclip. Give it the Paperclip GitHub repo as context, let it understand the API and architecture, and use it as your partner for configuring agents, troubleshooting issues, managing secrets, and planning VPS migrations.

As shown in the transcripts, experienced users find this invaluable because the Claude Code project understands the full Paperclip architecture, all the API endpoints, the heartbeat protocol, and can help you query and manage your companies programmatically. You can ask it things like "What companies do I have in Paperclip and what are they doing?" and it will list everything out.

Setting Up API Keys and Secrets

Paperclip has an internal system for storing environment variables and secrets, though there is no dedicated UI section for it in the dashboard. The recommended approach is to use your Claude Code helper to create and manage secrets programmatically. Your Claude Code partner can create the secrets, tell each agent where they live and how to use them, and ensure your agents can access external tools and services securely.

Remote Access with Tailscale

By default, Paperclip runs on localhost and is only accessible from your machine. For solo entrepreneurs who want to monitor their AI company from anywhere, Tailscale provides a simple solution for remote access without complex server configuration. When you are ready for a more permanent deployment, you can move Paperclip to a VPS and deploy to services like Vercel.

Start Small, Scale Gradually

The most successful Paperclip users consistently share the same advice: do not try to automate everything on day one. Start with one or two agents. Get them working reliably. Understand the heartbeat rhythm, the ticketing workflow, and the feedback loop. Then gradually add more agents and expand your organization.

As one practitioner put it: if you try to one-shot an entire startup with AI, it is super fun for the first 30 minutes and then falls apart. The key is building incrementally and encoding your standards into each agent's instruction files as you learn what works.

Common Issues and How to Fix Them

Agent runs heartbeat but reports "No work found" Check three things: (a) is the task assigned to the correct agent ID? (b) is the task status set to "todo"? (c) did you source the environment variables? Verify with echo $PAPERCLIP_AGENT_ID.

Agent starts but times out before completing The task is either too complex for the default timeout, or the API key has expired. Fix by increasing timeoutSec in agent configuration or verifying your API key on the Anthropic dashboard.

The CEO keeps hiring but never delegates work Check the CEO's heartbeat instructions. Make sure the checklist includes steps for breaking goals into projects and creating specific task assignments for team members.

Two agents working on the same task This should not happen if you are using Paperclip's checkout system correctly. Tasks are atomically checked out — only one agent can claim a task at a time. If you see duplication, verify that agents are using the checkout API call before starting work.

Dashboard shows $0 spend despite heavy usage This is expected if you are using a subscription plan (Claude Code Max, etc.). Token spend tracking only shows real dollar amounts when using pay-per-token API credits.

Previous instance still running If you get errors about port 3100 being in use, a previous Paperclip instance may still be active. Kill it with lsof -i :3100 to find the process, then terminate it.

General troubleshooting: Run pnpm paperclipai doctor --repair as your first step when anything goes wrong. It checks the database, configuration, permissions, and auto-repairs what it can.

Your LeadForge Company After Setup

When you complete this tutorial, here is what your LeadForge company looks like inside Paperclip:

Org Chart:

Board of Directors (You)
└── CEO Agent (Claude Opus 4.6)
    ├── Lead Research Specialist (Claude Sonnet 4.6)
    ├── Outreach Copywriter (Claude Sonnet 4.6)
    └── QA Specialist (Claude Sonnet 4.6)

Active Projects:

  • Client Prospect Research

  • Outreach Content

  • Quality Assurance

Recurring Routines:

  • Daily Lead Quality Report (QA, 10 AM daily)

  • Weekly Strategy Review (CEO, Monday 9 AM)

  • Daily Outreach Drafts (Copywriter, 8 AM daily)

Budget Controls:

  • Per-agent monthly limits with 80% warning and 100% hard stop

  • Circuit breaker for abnormal spending patterns

Skills Installed:

  • Company-wide Paperclip skills (auto-installed)

  • Role-specific skills for research, copywriting, and QA

Governance:

  • Board approval required for new hires

  • Full audit trail on every action

  • Immutable activity log with complete traceability

What Comes Next

Once your LeadForge company is running, you can expand in several directions:

Add Specialized Agents: Hire a social media agent to promote your lead generation services, a data analyst to track conversion metrics, or a sales agent to manage client communications.

Import Company Templates: Browse the Paperclip companies repository for pre-built teams you can import directly into your instance. Think of it as acquiring a proven team structure rather than building from scratch.

Set Up Multiple Companies: Use Paperclip's multi-company support to run separate operations. Maybe LeadForge handles B2B outreach while a second company handles content marketing — all from the same Paperclip dashboard with complete data isolation.

Move to a VPS: When you are ready for 24/7 operations, deploy Paperclip to a cloud server. Your agents will work around the clock — the researcher finding prospects at 2 AM, the copywriter drafting outreach at 4 AM, and the QA agent reviewing everything at 7 AM, all while you sleep.

Watch for Clipmart: The upcoming Clipmart marketplace will let you download entire pre-built company templates with one click. When it launches, you will be able to import a battle-tested lead generation operation and customize it rather than configuring every agent from zero.

The zero-human company starts with a single terminal command. Where it goes from there is up to you.

Parash Panta

Content Creator

Creating insightful content about web development, hosting, and digital innovation at Dplooy.