27.5k stars: orchestrate 20 AI agents like a company, not scripts

You know that feeling when you have 15 Claude Code terminals open, each working on something different, and you reboot and lose all context? Or when an agent gets stuck in a loop and burns $50 in tokens before you notice? Before Paperclip: you manually copy context between agents, can't track who's doing what, and have zero cost visibility until the API bill arrives. After: every task is a ticket, every agent has a budget, every decision is logged, and agents resume work after reboots.

Think of Paperclip like an operating system for AI companies. You run `npx paperclipai onboard` and it spins up a Node.js server with embedded Postgres. You define a company mission ('Build #1 AI note app to $1M MRR'), hire agents (CEO Claude, CTO Cursor, engineers), and set monthly budgets ($30-60/agent). Agents wake on heartbeats (every 4h, 8h, etc.), check their assigned tickets, do work, and report back. Every task traces up through goals to the mission — agents always know why they're doing something. You sit on the board: approve hires, review strategy, pause agents, adjust budgets. All from a mobile-ready dashboard.

• Bring Your Own Agent — why YOU care: works with Claude Code, OpenClaw, Cursor, Codex, or even bash scripts. If it can receive a heartbeat signal, it's hired. No vendor lock-in.
• Goal alignment — why YOU care: every task carries full goal ancestry up to the company mission. Your agents know what to do AND why — no more re-explaining context.
• Heartbeats — why YOU care: agents wake on schedule (every 4h, 8h, 24h), check work, and act. Recurring jobs like customer support or reports happen automatically without you remembering to kick them off.
• Cost control — why YOU care: monthly budgets per agent. When they hit 80%, you get warned. At 100%, they auto-pause. No runaway loops burning hundreds in tokens.
• Governance — why YOU care: you're the board. Agents can't hire new agents without approval. You can pause, resume, override, or terminate any agent at any time. Autonomy is a privilege you grant.
• Multi-company — why YOU care: one deployment runs unlimited companies with complete data isolation. Run 50 autonomous businesses from one control plane.
• Persistent state — why YOU care: agents resume the same task context across heartbeats and reboots. No more losing everything when your laptop restarts.

Who It Is For

If you're running 5+ AI agents and losing track of what each one is doing — this is for you. Perfect for solo founders building autonomous businesses, teams experimenting with multi-agent systems, or anyone who's hit $100+ in surprise API bills from runaway loops. Not useful if you only have one agent or just want a chatbot interface.

Worth Exploring?

The 27.5k stars aren't hype — this solves a real pain point for power users of AI agents. The architecture is solid (atomic task checkout, persistent state, goal-aware execution). The MIT license and self-hosted model mean no vendor lock-in. Try it if you have 5+ agents or want to build autonomous businesses. Skip if you're just experimenting with one Claude session.