Paperclip
Paperclip is an open-source Node.js platform that orchestrates teams of AI agents as employees of autonomous, self-hosted businesses.
Pros
- + Org-chart model with roles, reporting lines and governance workflows
- + Bring Your Own Agent — works with any runtime or HTTP bot
- + Per-agent monthly budgets with hard spending limits
- + Atomic task checkout prevents duplicate work
- + Multi-company support with strict data isolation
Cons
- − Conceptually heavy — the company metaphor is a learning curve
- − Production deployments need an external Postgres database
- − Self-hosted only; no managed offering
- − Autonomous-business framing is ambitious and largely unproven
What Is Paperclip
Paperclip is an open-source Node.js server with a React UI that orchestrates teams of AI agents to operate autonomous businesses. It leans fully into an organizational metaphor: individual agents are employees, they have roles and reporting lines, and the platform is the company that coordinates them.
The problem it tackles is the operational mess of running many agents at once. Without a control plane you end up with scattered agent tabs, manual context shuffling, disorganized configuration, runaway token costs and no clear record of what autonomous work actually happened. Paperclip consolidates all of that into a single management interface.
Architecture
Paperclip is a TypeScript codebase. It runs as a single Node.js server with an embedded PostgreSQL database for local use, or an external Postgres for production.
| Component | Detail |
|---|---|
| Control plane | Single Node.js process managing companies, agents and runs |
| Database | Embedded Postgres locally; external Postgres in production |
| Wakeup queue | Database-backed queue with coalescing for heartbeat execution |
| Triggers | Cron, webhook and API triggers for recurring routines |
| Isolation | Company-scoped entities; scoped API keys and run-specific JWTs |
Agents run on scheduled heartbeats and event-based triggers rather than staying persistently online. Each agent gets scoped API keys, run-specific JWTs and budget enforcement, and task checkout is atomic so two agents never pick up the same work. Multi-tenancy is handled by company-scoped entities — separate organizations on a shared deployment are fully isolated. The system supports encrypted secrets and object-storage backends, and every run is recorded in an immutable audit log.
Key Features
- Bring Your Own Agent — any runtime works: Claude Code, Codex, CLI tools or HTTP bots.
- Atomic task checkout — prevents double-work across concurrent agents.
- Per-agent budgets — monthly hard spending limits per agent.
- Multi-company — run several isolated organizations on one deployment.
- Governance — workflows that require board approval before action.
- Persistent state — agent state carries across heartbeats.
- Org structure — org charts, roles and hierarchical reporting lines.
- Audit logging — full, immutable activity traces.
What’s Missing
Paperclip’s company-as-software metaphor is powerful but heavy: the org-chart, roles and governance model is a real conceptual learning curve before you are productive. Production use requires you to provision and operate an external PostgreSQL database — the embedded one is for local development only. It is self-hosted with no managed tier, so infrastructure, upgrades and database operations are yours to run. Most importantly, the headline framing — agents operating autonomous businesses — is ambitious and largely unproven; the platform gives you the control plane, governance and budgets, but the business logic, the agents themselves and the judgment about what they should do are entirely things you bring. Anyone wanting a simple, single-agent assistant will find this overbuilt.
Who It’s For
Paperclip fits engineers and entrepreneurs who are deliberately running multiple agents as a coordinated team — with budgets, roles and approval gates — and who are comfortable operating a Node.js server and Postgres. It is overkill for one-off or single-agent use, and best suited to people experimenting seriously with agent-run operations.