Paperclip Orchestrates AI Agents into Zero-Human Companies
Paperclip, a free open-source dashboard, combines with Claude Code to manage proactive AI agents via heartbeats, budgets, and ticketing—eliminating the chaos of juggling 20+ terminals for autonomous business teams.
Why Paperclip Solves AI Agent Orchestration Chaos
Running multiple AI agents like those from Claude Code or OpenClaw often devolves into terminal hell: 20+ sessions scattered across windows, no central visibility, forgotten goals, and desynchronized progress. Paperclip fixes this with a unified dashboard that logs every task, conversation, and spend. Launched in early March 2025, it hit 36,000 GitHub stars in weeks under MIT license, trending over tools like OpenClaw by centralizing orchestration. One founder cited the pain: "every day he would have 20 different terminals running, 20 different Cloud Code sessions running. And I've definitely been there." Instead of reactive chats, agents operate via issues, approvals, and heartbeats—mimicking a real org chart where AIs hire AIs.
Tradeoff: Defaults to localhost, so deploy to a VPS for remote access. No built-in remote control out-of-box, but heartbeats (every 4-12 hours) keep agents proactive by refreshing context, checking tasks, and self-familiarizing—unlike stateless sessions that lose momentum.
Building Autonomous AI Companies from Scratch
Setup takes minutes: Clone from paperclip.ing, run one terminal command, name your company, and launch. Demoed with "Proofshot," a SaaS for AI-cleaned video testimonials: Users submit links, AI processes for embeddable widgets. Start with a CEO agent (Sonnet 3.5 Sonnet or custom models/providers per agent). Default first task: "Hire your first engineer and create a hiring plan."
CEO plans milestones—MVP backend, AI pipeline, frontend UI—then requests approval for hires like a full-stack engineer (integrations, design, DB, frontend). Approve via inbox; it spawns the agent with minimal instructions plus native Paperclip skills (API, heartbeat protocol). CEO delegates: Engineer gets 5 subtasks. Human acts as "board," issuing high-level issues like "Hire QA for bug checks, migrate tasks to GitHub project 'Initial MVP,' iterate until done."
"You have this entire company now. I have a CEO, I have a social media agent, I have a marketer that manages a copywriter, a strategist, a designer, and a researcher." This scales to sub-units: Author uses it for AI Automation Society (300k+ community) to automate content without firing humans—chip away at departments.
Projects sync to GitHub repos for code persistence. Multiple companies per instance: New ventures or departmental silos. Budgets per agent/month cap spend (zeros shown with subscriptions like Anthropic; Paperclip tokens track otherwise).
Agent Customization: Skills, Heartbeats, and Routines
Agents ship with Paperclip-native skills: Understand dashboard, issues, hires. Enhance via files—soul (persona: "You own the P&L. Default to action. Hold the long view while executing near-term."), heartbeat (execution checklist), tools, agents.md. Customize voice/tone, capabilities (task assignment, new hires).
Pull skills from skills.sh marketplace (free, audited but vet security): Frontend design guidelines, web design. Paste GitHub URL; auto-installs company-wide or per-agent. Author prepped a Claude Code project with Paperclip GitHub, X discussions, gotchas—acts as executive assistant for config, VPS migration, monitoring.
Heartbeats: Agents "wake up fresh" on schedule, review work, act. "They wake up with fresh context and fresh memory, which is why it's important that they have to check their work. They look at their tasks. They get familiarized with their environment first."
Routines (beta): Recurring workflows like nightly security scans—schedule via cron/webhooks/terminal, assign to QA/CEO. Tracks runs. Toggle board approval for hires to let CEO auto-scale.
Import pre-built companies (G-Stack, Agency Agents) like acquisitions—bring teams/frameworks instantly.
Interactions differ from pure Claude Code: Less conversational, more issue-based. "It's much less like having a long ongoing conversation, it's more so creating issues, leaving comments, you know, waking them up on a heartbeat."
Real-World Integration and Expansion Tactics
Pairs with Claude Code for execution (plans, commands, web fetches). Bring custom agents (OpenClaw, Cursor). Author runs 7 agents (CEO, marketer→copywriter/strategist/designer/researcher, social) for AIS content scaling—high-level goals, notifications, inbox approvals.
For existing biz: Automate subunits (e.g., LinkedIn strategy). New ideas: Standalone companies. Track via dashboard—tasks in progress, activity log, agent spends.
Pro tip: Feed Claude Code the Paperclip repo for meta-assistance. Evolve files organically; manual edits for tweaks.
Tradeoffs: Approval gates slow autonomy (toggle off for growth). Skills add power but risk (audit yourself). Local limits scale; VPS needed for 24/7. Heartbeats cost tokens but enable proactivity over one-shots.
Results: 30 minutes to CEO+team executing MVP milestones. No humans beyond board oversight.
Key Takeaways
- Run
curl -fsSL https://paperclip.ing/install | bashto launch dashboard instantly. - Start every company with CEO + hire task; approve to bootstrap team.
- Use heartbeats (4-12h) for always-on proactivity without constant prompting.
- Install skills.sh repos (e.g., frontend-design) for specialized capabilities.
- Sync projects to GitHub early for code QA and iteration.
- Set per-agent budgets to control costs before scaling.
- Prep a Claude Code assistant with Paperclip docs for setup troubleshooting.
- Import templates for instant teams; customize souls for role fit.
- Act as board: High-level issues only—let agents plan/hire.
- Deploy to VPS for remote access; localhost suits experiments.