How to Use Paperclip AI to Build a Zero Person Business

**TL;DR:** Paperclip is an open-source multi-agent orchestration system that lets you run a 24/7 “AI company” on a VPS; the video shows how to deploy it on Hostinger, hire CEO/research/writing/editing agents, and chain tasks via Issues so content production runs while you sleep. ## What Paperclip Is and Why It Exploded Paperclip became one of the fastest-growing open-source AI projects of 2026, hitting 38k GitHub stars in three weeks. Instead of prompting a single model, you “hire” a whole org-chart of agents—each with a role, budget, and shared task board—inside one dashboard. The mental model flips from “craft the perfect prompt” to “sit on the board”; a CEO agent keeps the lights on, delegates to specialists, and every action is logged as a GitHub-style Issue. ## How Multi-Agent Orchestration Works Single agents (OpenClaw, Claude Code, etc.) already write code, browse, email, and loop until a job is done. Paperclip adds orchestration: * **CEO agent** – breaks annual goals into weekly tasks, decides who gets hired next, monitors burn-rate. * **Research agent** – scrapes news, writes briefings. * **Writing agent** – turns briefings into articles. * **Editor agent** – fact-checks, tone-matches, formats. Every agent has its own persona, skills list, and model adapter. You can hot-swap models: Claude Opus for heavy reasoning, Gemini Flash for cheap formatting, GPT-4 for legacy code. Token spend is tracked per agent; budgets can be capped and topped-up like prepaid phones. ## Deploying Paperclip in the Cloud (Zero-Downtime) Running locally means agents die when your laptop sleeps. A 5-dollar VPS keeps them alive. The video walks through Hostinger’s one-click image, but any KVM box works. ### 1. Pick a VPS tier KVM2 (2 vCPU, 8 GB RAM) is the sweet spot; KVM1 chokes when three agents browse simultaneously. ### 2. Claim the coupon Use the creator’s link; the coupon only applies to 12-month or longer renewals. ### 3. Add daily backups Costs pennies and lets you roll back if an agent forks your file system. ### 4. Choose the closest data-center Let Hostinger’s latency tester decide; agents hammer APIs all day, every millisecond counts. ### 5. Inject API keys Paste at least one key—Anthropic, OpenAI, Gemini, or Cursor—before hitting Deploy. The installer spins up Docker, Postgres, and the Paperclip UI; the purple “Open” button appears after 5–10 min. ## First Login: Founding Your AI Company 1. Name the company (e.g., “ContentEngine”). 2. Describe the mission: “Track AI news and ship one polished article per week.” 3. Create the CEO agent * Name: CEO John * Adapter: Claw Code * Model: Claude Opus 4.6 * Prompt: default “run the company” template. 4. Launch. Within seconds the Inbox shows “Request to hire Research-Agent-001”; approve and the second agent appears. Dashboard tabs: Dashboard – burn-rate, active Issues, token quota. Inbox – approval queue for hiring, budget increases, new routines. Issues – Kanban board where agents open, assign, and close tasks. Goals – high-level OKRs the CEO iterates on. Agents – skill editor, budget caps, model swapper. Settings – auth, backups, telemetry toggle. ## Building a 3-Person Content Team in 5 Minutes Issue-driven workflow keeps humans in the loop only when they want to be. 1. Open Issue: “Hire research agent – duties: find 5 AI highlights weekly, write briefing, hand off to writing agent.” Assignee: CEO → auto-approved → Research-Agent-002 is spawned. 2. Repeat for Writing-Agent and Editor-Agent; paste brand-voice guidelines into each agent’s Instructions field. 3. Create production Issue: “Week-1 content batch.” Description spells out the chain: * Research finds 5 topics → drops briefing in Issue #42 → @mentions Writing-Agent. * Writing-Agent produces draft → drops markdown → @mentions Editor-Agent. * Editor approves or re-opens; when closed, CEO is notified and the cycle repeats next week. All state lives in Issues; nothing is hidden in prompt memory, so you can audit every decision and roll back by re-opening an Issue. ## Power-Ups: Skills, Routines, Budget Guardrails **Skills** Browse skills.sh, copy YAML snippets (e.g., “scrape Hacker News”, “generate DALL·E hero image”). In Paperclip → Skills → Add, then tell the CEO which agent should learn the skill. Agents auto-import the function into their sandbox. **Routines** Sidebar → Routines → Create. Example: “Every Monday 08:00 UTC open Issue ‘Research weekly AI highlights’ and assign to Research-Agent.” Cron-like scheduler fires even if you’re asleep; routines can also trigger budget reports or backup jobs. **Budget** Per-agent monthly cap, 80 % email warning, 100 % hard stop. Prevents a runaway browsing loop from burning your Anthropic balance overnight. ## Caveats and Pro Tips * Paperclip treats AIs like people—great for realism, but also brings human problems: vague specs, hand-off friction, interpretation drift. * Write persona and acceptance criteria as if onboarding a junior contractor; ambiguous verbs == endless loops. * The project is three weeks old; expect UI glitches and occasional 502 on rapid agent spawning. * For single-shot tasks, a well-prompted lone agent can outperform an un-tuned multi-agent crew. Use Paperclip when the workflow is recurring and worth automating. ## Next Steps Your content engine now runs unattended on a Hostinger VPS: research → write → edit → publish, tracked and billed in one pane of glass. Clone the setup, swap the skills, and you’ve got a support desk, a coding squad, or a market-research firm—still zero human employees. **Source:** [How to Use Paperclip AI to Build a Zero Person Business – Youri van Hofwegen](https://www.youtube.com/watch?v=4Bra0p4lsac)