# Paperclip (AI Agent Orchestration) Paperclip is an open-source app for running a team of [[AI Agents]] the way you would run a company. You define a goal, "hire" agents into an org chart (CEO, CTO, engineers, marketers...), give each one a monthly budget, and approve their strategy before they start. Their tagline sums it up: "If OpenClaw is an *employee*, Paperclip is the *company*." It's built by Paperclip Labs (paperclip.ing), not to be confused with the biomedical research CLI of the same name in [[Paperclip]]. Why does this exist? Once you run more than a couple of agents, the bottleneck stops being what one agent can do. It becomes coordination: who is working on what, why, at what cost, and who signed off. Their README describes the pain well: 20 [[Claude Code]] terminals open, no idea which one does what, and everything lost on reboot. Paperclip's answer is to reuse the machinery companies built for people (reporting lines, approvals, budgets, an audit trail) and apply it to agents. The model has a few moving parts: - **Bring your own agent**: Paperclip doesn't run models itself. Adapters connect [[Claude Code]], [[OpenAI Codex]], [[Cursor]], [[Gemini CLI]], [[OpenCode]], [[Hermes Agent]], [[OpenClaw]], [[Kimi Code]], shell scripts or HTTP webhooks. Their rule: "If it can receive a heartbeat, it's hired." - **Heartbeats**: agents don't run all the time. They wake up on a schedule or on an event (task assigned, new comment, blocker cleared), act, and go back to sleep. Delegation flows up and down the org chart - **Goal alignment**: every task links back through project and company goals to the mission, so the agent gets the "why" along with the "what" - **Tickets**: you talk to agents through task threads. Tool calls and decisions are traced, and the audit log is append-only - **Budgets**: per agent, project and company. Soft warning at 80%, auto-pause at 100%, and you can override - **Governance**: you act as the board. Agents can't hire other agents without your approval, and you can pause, reassign or terminate any of them at any time - **Multi-company**: one deployment can host many companies with isolated data Under the hood it's a [[Node.js]] server with a React UI. Locally, it sets up an embedded [[PostgreSQL]] database; you can point it at your own Postgres or deploy it remotely. Setup is one command (`npx paperclipai onboard --yes`) and no Paperclip account is needed. It's [[MIT License|MIT-licensed]] and [[Self-hosting|self-hosted]]; a hosted cloud version is waitlist-only. The repository was created in March 2026 and crossed 96K GitHub stars by October 2026, with releases versioned by date (v2026.1001.0 shipped on October 1, 2026). Early on, the project pitched itself as "open-source orchestration for zero-human companies". The current wording is softer: "the app people use to manage AI agents for work". I like the mental model a lot. A company is a well-understood way to coordinate many workers with partial context, and agents need exactly that (it's the same reason my own vault agents have a receptionist, handoff rules and chain limits). The closest alternative I know is [[Multica]], which frames the same idea as project management for mixed human and agent teams. One thing to keep in mind though: Paperclip's governance covers what agents do *inside Paperclip* (hiring, strategy, budgets). Their FAQ is explicit that securing the agents themselves is on you. And since v2026.1001.0, execution harnesses default to full auto (e.g., Claude runs with approve-all, Codex bypasses its sandbox approvals) unless you configure a restrictive mode explicitly. So the "board of directors" doesn't make an unsandboxed coding agent safe on your machine. Budgets also rely on reported spend; the README notes that in-flight work can delay a stop. ## References - Website: https://paperclip.ing/ - About: https://paperclip.ing/about/ - Product overview: https://paperclip.ing/product/ - Open source page: https://paperclip.ing/product/open-source/ - Heartbeats: https://paperclip.ing/product/heartbeats/ - Changelog v2026.1001.0: https://paperclip.ing/changelog/v2026.1001.0/ - LLM-readable summary: https://paperclip.ing/llms.txt - Documentation: https://docs.paperclip.ing - GitHub repository and README: https://github.com/paperclipai/paperclip - Hermes Agent adapter (Nous Research): https://github.com/NousResearch/hermes-paperclip-adapter - Hacker News submissions (March 2026, "zero-human companies" tagline): https://news.ycombinator.com/item?id=47287414 and https://news.ycombinator.com/item?id=47277096 ## Related - [[AI Agent Orchestration]] - [[Multi-Agent System (MAS)]] - [[Multica]] - [[Gas Town]] - [[Claude Code Agent Teams]] - [[Human-in-the-Loop]] - [[Open Source]]