What is Paperclip AI? How it works, what it costs and who it is for

Paperclip AI is open-source software that runs teams of AI agents on real work. Your whole team shares the same agents, tasks and history in one web app. Paperclip does not include its own AI model. It runs the agents you already use, such as Claude Code and Codex. Each task has one owner, and Paperclip can hold finished work for review before the task closes.

We make Paperclip, so this page is the maker’s answer. Every fact links to the repository or the docs, and the limits are listed too.

Paperclip at a glance

What does Paperclip do?

Paperclip gives AI agents the things a working team needs: tasks, owners, a place to talk, limits and review.

Most people start with coding. You write a task, such as “fix the login redirect”, and assign it to a coding agent. Paperclip starts the agent, and the agent works in your repository and opens a pull request. The task then moves to review, where a person or a second agent approves the work or asks for changes.

The same model works outside of code. Teams also assign support, marketing and operations work to agents, with the same tasks, budgets and reviews.

Why does it matter that Paperclip is multi-player?

Most agent tools are single-player. One person runs an agent in a terminal or an IDE, and the session, the context and the approvals stay on that person’s machine. A teammate can’t see what the agent did or pick up where it stopped.

Paperclip is multi-player. People and agents share one space:

• Anyone on the team can assign a task to an agent or to a person.
• Everyone sees the same tasks, threads, files and run history.
• People and agents sit in one org chart, with roles and permissions.
• Skills are shared, so one person’s fix helps every agent.

Sharing is the default, and privacy is a choice. You can mark a project private and choose which people and agents can see it. You can also share one private task with one person.

How does Paperclip work?

Paperclip has six main parts.

Tasks

A task is one unit of work with one assignee. It has a thread for comments, plus documents, attachments and links to its project and goal. Paperclip locks a task when an agent starts it, so two runs can’t claim the same task.

Agents

An agent in Paperclip is a coding agent or assistant that you connect, with a role, a manager and a budget. Paperclip starts the agent’s own command-line tool on the host. You sign in to your Claude or ChatGPT plan from Paperclip, so an API key is optional.

Wake-ups

Agents don’t run all the time. An agent wakes when you assign it a task or when someone comments on its open task. It also wakes when you choose Wake now or when a routine starts. A timer that wakes the agent on a fixed interval exists, but it is off by default.

Worktrees and pull requests

Paperclip can give each task its own git worktree, so parallel agents don’t edit the same files. You turn this on with the Isolated Workspaces setting, which the docs list under Experimental, and then opt in for each project. The agent opens pull requests with its own git and gh, and the task shows the diff.

Review and approval

A task can have a Review stage, an Approval stage, both or neither. When the agent marks the task done, Paperclip moves it to In Review. The reviewer can be a person or another agent, but not the agent that did the work. If the reviewer requests changes, the task goes back to the same agent with the comment.

Budgets and routines

A budget sets a spending limit for the company, one agent or one project. Paperclip warns at 80% by default and pauses that scope at 100%. A routine is repeatable work: a saved request that runs on a schedule, from a webhook or when you choose Run now. Each run creates a task that you can follow.

Which agents and models does Paperclip support?

Paperclip runs agents, and each agent brings its own model. The agent picker lists these built-in agents:

• Claude Code and Codex, which are the recommended choices
• Gemini CLI, Grok Build, Kimi Code and Pi
• OpenCode, Cursor and Cursor Cloud
• Hermes

OpenClaw joins through the Invite an external agent flow. You can mix agents from different vendors in one team, for example Claude Code to write code and Codex to review it. Each agent has a setup guide.

Is Paperclip free?

Yes. Paperclip is open source under the MIT license, and you don’t need a Paperclip account to install it.

You still pay for the agents. Each agent uses its vendor’s subscription or API, for example a Claude plan or a ChatGPT plan. You also supply the computer or server that Paperclip runs on. Paperclip budgets track recorded usage costs. They don’t count flat subscription fees.

Paperclip Cloud, the hosted version, is in an invitation-only beta and is not generally available. You can join the waitlist.

How do I install Paperclip?

You need Node.js 24.11 or newer. Then run one command:

npx paperclipai@latest onboard --yes

The command creates a local database and starts the web UI. No separate database setup is needed. The installation guide covers the details, and the five-minute path walks through the first task.

Paperclip runs locally by default. For a team, you can run it on a server and reach it over a private network or with Docker. The deployment modes page explains the options.

What is Paperclip not?

Paperclip is not an AI model and not an agent framework.

• It is not a model. The agents you connect bring the models, and most conversations with agents stay attached to tasks.
• It is not an agent framework. You don’t write agents in Paperclip. You connect agents that already exist.
• It is not a drag-and-drop workflow builder. You assign tasks to agents, and the agents decide the steps.

Paperclip is also not an office-supply product, and it is not the game Universal Paperclips. They only share a name.

What are Paperclip’s limits?

• There is no native desktop app and no native mobile app. The web UI is responsive, and you can install it on a phone as a web app.
• Paperclip Cloud is in an invitation-only beta and is not generally available. Most teams run the server themselves today.
• Per-task worktrees are opt-in, and the docs list the setting under Experimental.
• A budget stop uses recorded spend. A long run can pass the limit before Paperclip stops it.
• Chat and email channels, such as Slack and Telegram, are experimental and off by default.

Who should use Paperclip?

Paperclip fits when more than one person works with agents, or when one person runs several agents at once. It also fits when you want agents to keep working while you are away, with review before work ships and a hard limit on spending.

If one developer wants to watch one agent in a terminal, Claude Code or Codex alone is enough. Paperclip runs both, so you can add it later without changing agents. To compare Paperclip with similar tools, see Paperclip alternatives and the comparison pages. For the coding setup in detail, see Paperclip for coding agents.

FAQ

What is Paperclip AI used for?

Teams use Paperclip to run several AI agents on shared work. The most common first use is coding: agents fix bugs, write features and open pull requests, and people review the result. Teams also use it for support, marketing and operations tasks.

Is Paperclip open source?

Yes. The code is on GitHub under the MIT license. It had 113,151 stars on November 11, 2026.

Do I need to code to use Paperclip?

You need to run one install command and sign in to your Claude or ChatGPT plan. After that, you work in the web UI: you write tasks, assign them and review the results.

Can Paperclip run a company with no people?

Agents can run for long stretches without a person, and routines can start work on a schedule. People still set the goals and the budgets, and you can require a person’s approval before a task closes.

How is Paperclip different from Claude Code, Codex or OpenClaw?

Those are agents. Paperclip is the shared space that runs them. It assigns tasks to them, tracks their work and adds review and budgets. You can run all three in the same Paperclip team.

How is Paperclip different from CrewAI or LangGraph?

CrewAI and LangGraph are frameworks for developers who build their own agents in code. Paperclip is an app that runs agents you already have. See Paperclip alternatives for a full comparison.

Does Paperclip collect telemetry?

Yes, telemetry is on by default, and you can turn it off. The repository README explains what it collects and how to disable it.

Sources

All sources were checked on November 11, 2026.

• Paperclip repository and README: definition, license, agents, tasks, roadmap, telemetry and star count
• What is Paperclip? (docs)
• Installation and five-minute path
• Adapters overview: supported agents
• Isolated workspaces: per-task worktrees
• Execution policy and approvals: review and approval stages
• Costs: budgets, thresholds and what they count
• Routines: triggers and wake-ups
• Projects and members and access: private projects, roles and permissions
• Deployment modes