Paperclip is an open-source agent-orchestration workspace: org charts, goals, budgets, governance, and task threads for a team of AI agents from different runtimes (OpenClaw, Codex, Claude Code, Cursor, Gemini CLI, OpenCode, Pi, Hermes). 'If it can receive a heartbeat, it's hired.' Agents wake for work, follow-up messages, or schedules; delegation flows up and down the chart; spend is tracked and capped at company, agent, and project level.
What it is
Paperclip (paperclip.ing, repo paperclipai/paperclip, MIT, TypeScript at roughly 96,000 GitHub stars with 6,253 open issues at a very active pace) is a Node.js server plus React UI that orchestrates a team of AI agents to run a business. The pitch in the README’s own words: “If OpenClaw is an employee, Paperclip is the company.” You define a goal (“Build the #1 AI note-taking app to $1M MRR”), hire a team - CEO, CTO, engineers, designers, marketers - as bots from any provider, review the strategy, set budgets, and hit go. It looks like a task manager; under the hood it is org charts, budgets, governance, goal alignment, and agent coordination.
The feature set that makes it a workplace
The org chart gives every agent a boss, a title, reporting lines, scoped secrets, and connection permissions; humans and agents sit on the same chart. Heartbeats wake agents for assigned work, follow-up messages, or configured schedules, and delegation flows up and down the chart. Cost control operates at company, agent, and project level with threshold alerts and automatic pause at configured limits - the budget line is enforced, not suggested. Governance adds review and approval stages (approve hires, pause, reassign, stop work). Task threads keep conversations, plans, blockers, files, and run history attached to the work; artifacts get previews and passage-level comments; shared skills ship with a Skill Studio for testing saved inputs and restoring versions; scheduled routines run recurring work via schedule, API, or webhook. One deployment can host multiple organizations with separate agents, permissions, and histories, and there is a catalog of ready-made teams.
Install and run
Self-hosted needs no Paperclip account: with Node.js 24.11+, npx paperclipai@latest onboard --yes configures and persists everything locally (default bind is trusted loopback; --bind lan or --bind tailnet for the authenticated presets). npx paperclipai test-drive runs an isolated foreground instance pre-initialized with a CEO agent (harness flags select codex or opencode with your own keys), configure edits settings. From source: clone, pnpm install && pnpm dev, which serves the UI and API on localhost:3100 with an embedded auto-created PostgreSQL database. A managed Paperclip Cloud is behind a waitlist.
Where it fits for local AI
Paperclip is the management layer, not another harness: agents keep their own runtimes and models (Claude Code, Codex, OpenClaw, Pi, Hermes, Gemini CLI, Cursor, OpenCode), and Paperclip supplies the org structure, the budget enforcement, and the governance layer that turning agents loose on business goals actually requires. Local-AI angle: mixed fleets are explicitly in scope - OpenClaw on your own hardware running open weights can be “hired” next to a Claude Code agent, with Paperclip tracking whose heartbeat, whose budget, and whose approvals. The six-thousand-open-issues pace buys features fast but means expecting sharp edges; self-host least-privilege as with any agent platform that can spend money and call external APIs.
Sources: GitHub repo · paperclip.ing · docs · installation guide