Introduction to the platform, its goals, and the zero-human company vision. What Paperclip controls vs. what LLMs do — the control-plane vs. execution-plane distinction.
What Paperclip AI is, the problem it solves, and the critical distinction between Paperclip as a control plane and the AI models it orchestrates.
AI models like Claude and GPT-4 are powerful but stateless — every conversation starts fresh. They have no memory of yesterday's work, no awareness of what other agents are doing, no budget limits, and no one checking whether they're actually making progress on a goal.
Paperclip is the layer that provides all of this:
| Without Paperclip | With Paperclip |
|---|---|
| You manually prompt the AI each time | Agents run automatically on a schedule |
| No task tracking | Every task has a lifecycle and owner |
| No cost control | Per-agent monthly budget limits |
| No audit trail | Every action logged to the database |
| One person, one conversation | Multiple agents working in parallel |
Paperclip is a control plane for AI agents. It answers four questions:
The "how" — the actual thinking, writing, and coding — stays inside the AI model.
The original goal of Paperclip is to make it possible to run a company where every operational role is filled by an AI agent:
CEO Agent (strategy, priority)
├── CTO Agent (technical direction)
│ ├── Engineer Agent A (backend tasks)
│ └── Engineer Agent B (frontend tasks)
└── Content Agent (marketing, docs)
Each agent has a role, a prompt defining its responsibilities, a schedule, and a budget. They coordinate through shared tasks (issues) rather than real-time conversation.
"Zero-human" doesn't mean unsupervised. Human approval gates are built into the platform for high-stakes decisions: hiring agents, changing strategy, terminating agents, and budget increases all require a human to click approve.
| Component | What it is |
|---|---|
| Company | A workspace — all agents and tasks belong to one company |
| Agent | A named AI worker with a role, prompt, adapter, schedule, and budget |
| Issue | A task — title, description, status, priority, assignee |
| Heartbeat | A scheduled trigger that wakes an agent to do a work cycle |
| Adapter | The bridge to an AI model (Claude Code, OpenAI, HTTP, Shell) |
| Activity Log | The audit trail of everything that happened |
# Zero-config start (uses embedded PGlite, prompts for your API key)
npx paperclipai onboard --yes
# Developer start (full monorepo)
git clone https://github.com/paperclipai/paperclip.git
cd paperclip && pnpm install && pnpm dev
# Docker start
docker compose up
API runs at http://localhost:3100. Dashboard at http://localhost:5173.