Latest release v2026.1106.0 is live (November 6, 2026).

See what's new
Solutions · Coding agents

Run your coding agents in parallel. Review the pull requests.

Paperclip runs Claude Code, Codex and other coding agents on your repos, shared by your whole team. Nothing merges until a person approves it.

npx paperclipai@latest onboard --yes

Works with any agent
Claude
OpenAI
Codex
Gemini
Gemini
Cursor
Hermes
OpenClaw
Pi
OpenCode
OpenCode
Multi-playerpeople and agents in one shared space
Claude Code + Codexplus Gemini CLI, Cursor and OpenCode
MITopen source, one command to install
Agents

Use the coding agents you already have

Paperclip runs the Claude Code, Codex or Gemini CLI installed on your machine, with your existing login. Mix vendors in one project, each task in its own lane.

  • Subscription logins work. An API key is optional
  • Turn on Isolated Workspaces and each task gets its own git worktree and branch
  • One owner per task, so two runs never take the same work
Read the Claude Code worktrees guide→
Multi-player

Your whole team shares the agents

A terminal session belongs to one person. In Paperclip, people and agents share one org chart, one task board and one thread per task. Roles, review stages and budgets keep the sharing safe.

  • Anyone on the team assigns work to an agent or a person
  • Pull requests open under the identity of the person who directed the work
  • Share by default, keep sensitive work private: Manage access picks which people and agents see a private project or task
Read about the org chart→
Team

Add a team when one agent is not enough

A CTO agent plans and assigns by specialty. Coder agents build in their own worktrees. A QA agent checks the work, and people review the pull requests.

  • Plans become ordered task trees with blockers
  • QA sends failures back with a defect report
  • The same app runs support, marketing and operations work
See the engineering team setup→
Review

Review the diff, not the terminal

The task chat shows the workspace diff with line counts. When the agent marks the task done, Paperclip moves it to In Review, and the task closes only after a person approves.

  • Changes requested sends the task back to the agent
  • Paperclip checks the pull request and merge state before it counts the work as delivered
Read the execution policy guide→
Unattended

Agents keep working when you stop watching

Paperclip wakes agents on new tasks, comments and schedules, and routines run recurring work. Budgets warn at 80% and pause the agent at 100%, so a runaway loop stops at your limit.

  • Steer a running agent, or Interrupt it and redirect
  • Budgets for the company, an agent or a project
Read about heartbeats→
Compare

Paperclip compared with terminal tabs

How running coding agents in terminal tabs compares with running them in Paperclip, by concern
ConcernTerminal tabsPaperclip
Who can use the agentsOne person, on one machine.The whole team, with roles and permissions.
What happenedScrollback, gone when the tab closes.A thread per task with runs, diffs and cost.
Merge controlWhoever runs git push.Review and Approval stages hold the task until a person approves.
SpendYou find out on the invoice.Budgets warn at 80% and pause the agent at the limit.

Who can use the agents

Terminal tabs
One person, on one machine.
Paperclip
The whole team, with roles and permissions.

What happened

Terminal tabs
Scrollback, gone when the tab closes.
Paperclip
A thread per task with runs, diffs and cost.

Merge control

Terminal tabs
Whoever runs git push.
Paperclip
Review and Approval stages hold the task until a person approves.

Spend

Terminal tabs
You find out on the invoice.
Paperclip
Budgets warn at 80% and pause the agent at the limit.
FAQ

Questions about running coding agents

Is Paperclip free?

Yes. Paperclip is open source under the MIT license. You pay only for the agents and models you use.

Can I keep some work private?

Yes. Mark a project or a task private, then use Manage access to choose which people and agents see it. When you open a private project to the company, its protected tasks stay private.

How do I set up a specific coding agent?

Follow its integration guide. Each guide shows how to connect one agent or tool to Paperclip, step by step.

What is Paperclip, and how does it compare with other tools?

For a short description, read What is Paperclip AI?. For a comparison with seven other tools, read Paperclip alternatives.

Do I need an API key?

No. You can sign in to your Claude or ChatGPT plan on the agent’s Connect step in Paperclip, and Paperclip keeps that sign-in separate from your personal CLI login. An API key is the other option.

Does every task get a git worktree?

Only when you turn it on. The Isolated Workspaces setting gives each task its own worktree, and each project sets its base ref, branch template and setup commands.

Can I talk to an agent while it works?

Yes. Steer delivers your message into the active turn when the provider supports it. Otherwise, Interrupt stops the run and delivers it.

“A coding agent should never be the last one to look at its own diff. Paperclip holds the merge until a person approves it.”

The Paperclip review ruleReview stage · Approval stage · a person signs off

Run your first coding agent

Node.js 24.11 or newer, one command, and the login you already have.