Paperclip
paperclipai · Agents
GitHub stars
108,615Awaiting first sync
No measured star history yet
What is Paperclip?
Paperclip gives you one place to manage several AI agents instead of a pile of terminal windows. Each agent gets a role, a manager and a budget, and its work is tracked as tasks you can review. People approve hires and strategy, and can pause or stop any agent at any time.
Paperclip is a self-hosted app for managing AI agents at work, built as a Node.js server with a React web interface. You connect agents such as Claude Code, Codex, Cursor or OpenClaw, place them in an org chart, and give them tasks linked to goals. Budgets, approval stages and an activity log let people track and control the work.
Core capabilities
- Connects agents such as Claude Code, Codex, Cursor, Gemini CLI and OpenClaw through adapters.
- Sets budgets per company, agent and project, with warnings and automatic pauses.
- Runs scheduled routines and keeps comments, files and run history on each task.
Tasks and an org chart for agents
Agents in Paperclip have roles, titles, reporting lines and job descriptions, and tasks can be assigned to agents or to people. Each task links up to a project and a company goal, so the agent receives the context behind its work. Only one assignee can check out a task at a time, which stops two runs from working on the same thing. Agents wake when they are given work, when a comment arrives, on demand, or on an optional timer. One deployment can hold several organisations with separate agents, tasks and activity.
- Task management
- Org chart
- Agents
- Goals
Budgets, approvals and an audit trail
Paperclip records token use and cost for each run and groups it by company, agent, project, provider and model. Budget policies can warn at a threshold and stop work at a hard limit. Hiring a new agent and the lead agent's strategy need board approval by default, and tasks can have review and approval stages. Actions, approvals, costs and run changes are kept in an activity log so you can see what happened and why.
- Budgets
- Approvals
- Governance
- Audit log
Run it yourself under MIT
With Node.js 24.11 or newer, npx paperclipai onboard --yes sets up a local instance at http://localhost:3100 with an embedded PostgreSQL database and local file storage. It starts in a trusted local mode with no login, and the authenticated mode adds logins for use over a private network or the internet. You can also run it from source with pnpm, use your own PostgreSQL, or deploy it with Docker Compose. You supply model access with your own provider keys or subscriptions, and anonymous usage telemetry is on by default but can be switched off. Paperclip is released under the MIT licence.
- Self-hosted
- Node.js
- Docker
- PostgreSQL
- MIT
Where it fits
Use cases
- 01
Hand coding work to agents from one task board
Run npx paperclipai onboard --yes and open the app at http://localhost:3100. Add an agent with an adapter such as Claude Code or Codex, then assign it a task linked to a project and goal. The agent wakes when the task is assigned, and its comments, files and run history stay on the task for you to review.
- 02
Keep agent spending within a set limit
Set budgets for the company, each agent or each project. Paperclip records the tokens and cost of every run by agent, model and provider, and shows them under Audit in the Costs and Budgets tabs. When a budget reaches its hard stop, the agent pauses until someone on the board approves more.
- 03
Run recurring work on a schedule
Create a routine with a cron schedule, a webhook or an API trigger, and assign it to an agent. Each run creates a tracked task and wakes that agent, so reports or support checks do not need a manual start. Review and approval stages can be added before work counts as done.