Paperclip

What it is

A free, MIT-licensed Node.js server and React UI that orchestrates a team of AI agents. Per its README, it looks like a task manager but has an org chart, budgets, governance and goal alignment under the hood: your agents get a boss, a title and a job description, you set a company goal and "hire the team" (CEO, CTO, engineers, designers, marketers), and you approve the strategy and watch spend from a dashboard. The README says it is not an agent framework or a prompt manager: agents bring their own prompts, models and runtimes. The model itself is not included.

Setup

1. Install Node.js 24.11 or newer, plus your agent's CLI and account (Claude Code or Codex) or an API key from the model provider. The managed install is documented for macOS, Linux or WSL2; native Windows is not documented, so on Windows the documented route is WSL2 (not tested by Axo). Local Claude and Codex subscription sign-in also needs Python 3 and the provider CLI.

2. Run the onboarding. It starts the app with an embedded PostgreSQL database, no setup needed. The UI and API open on your own machine, on port 3100.

   npx paperclipai@latest onboard --yesCopy

3. In the UI, create your company, define the goal and create a CEO agent with its adapter (Claude Code, Codex and others).

4. Set budgets for the company and for each agent before you turn heartbeats on, then approve the CEO's strategy. The CEO cannot move tasks to in-progress until you approve it.

5. To start it again later:

   npx paperclipai runCopy

6. For an isolated trial with a CEO already created (set your own key first):

   ANTHROPIC_API_KEY=... npx paperclipai test-driveCopy

Cheat sheet

• Repository and README, docs and the five-minute guide.
• It costs tokens. The app is free, but every agent runs on your provider account. The official guide says: "Plan on spending $5-20 to play with the product, $20-100/month for an active company."
• Budgets exist at company, agent and project level. Company and agent budgets are monthly; a project budget is a lifetime cap. At 100% the agent is paused automatically, but the README says "usage reporting and in-flight work can delay a stop", so start low.
• You are "the board": hire approvals are optional and the CEO's strategy needs your approval. This is a governance board, not a Kanban board.
• Agents wake in heartbeats (on a timer you turn on, an assignment, a routine or a manual invoke; the README says a mention alone does not wake an agent), not continuously. Mutating actions, approvals, comments and cost events are recorded as activity.
• Cloud is only a waitlist, with no published price. Native Windows is not documented. Telemetry is anonymous and on by default; DO_NOT_TRACK=1 turns it off.
• Axo's note: skip it if you only use one agent for small jobs (the README says one agent is fine to start).
• Not verified by Axo: how it runs on Windows, and what a real team of agents costs.

Axo's take

AXO

Haven't run it myself: this page comes from the project's README and docs. Try it if you already run several coding agents and keep losing track of who is doing what and how many tokens they burn. Set a budget first: the app is free, your tokens are not.