Paperclip AI: Open-Source Agent Team Manager (2026)

Paperclip is a free, open-source (MIT), self-hosted app that organizes multiple AI agents — OpenClaw, Claude Code, Codex, Cursor, and more — into a single "company," complete with an org chart, budgets, and human approval gates. It's a Node.js server with a React UI that looks like a task manager on the surface, but underneath runs org charts, budgets, governance, and goal alignment.

It's built by paperclipai, a GitHub organization founded on March 2, 2026. As of this writing (September 28, 2026), the repository has roughly 91,765 stars and 15,816 forks, and it ranked #2 on GitHub's weekly trending list with +7,364 stars in a single week (#3 was Hindsight, an agent memory layer). The latest release is v2026.916.1, shipped 2026-09-21. This article covers Paperclip's main features, installation, basic usage, and how it compares to existing tools, based only on the official README and GitHub repository.

The README frames Paperclip's position with the line: "If OpenClaw is an employee, Paperclip is the company." It isn't a tool to make one agent smarter — it's a control plane for running many agents, across any provider or runtime, as a single organization. The basic flow is described in three steps.

The README argues that four things have to work for an organization of AI agents to actually produce: the tasks, the org, the training, and the infrastructure. Paperclip is built around exactly those four pillars.

Three points the README calls out specifically: atomic execution (task checkout and budget enforcement are atomic, so there's no double-work and no runaway spend), persistent agent state (agents resume the same task context across heartbeats instead of restarting from scratch), and governance with rollback (approval gates are enforced, config changes are revisioned, and bad changes can be rolled back safely).

The official installer is the recommended path — download the checksum-verified shell script and run it.

The installer ensures Node.js 24.11+ is available, installs a managed Paperclip CLI under ~/.paperclip/cli, and starts interactive onboarding; it can also run as a background service on supported Linux and macOS systems. To try it without a permanent install, use the npx onboarding command.

For an isolated instance already initialized with a CEO agent, use test-drive. It stays in the foreground, never installs a service or creates a first task, and only opens the browser once setup succeeds.

For a manual setup, clone the repository and run pnpm install && pnpm dev. This starts the API server at http://localhost:3100, with an embedded PostgreSQL database created automatically.

- Declare a goal: write down what you want to achieve, as text
- Hire agents: assign Claude Code, Codex, Cursor, OpenClaw, or any agent you like to roles such as CEO, CTO, or engineer (bring your own agent)
- Approve the strategy: review what the CEO agent proposes and set budget caps
- Hit go: agents wake on their heartbeat schedule and work through ticketed tasks
- Monitor from the dashboard or your phone: check diffs, screenshots, and test results, pausing or approving as needed

Recurring work — customer support, social posts, reports — can be automated with Routines, triggered by cron, webhook, or API calls, so nothing needs a manual kickoff each time. Plugins let you extend Paperclip with new tools and UI without forking it. For related agent-infrastructure background, see the Claude Code Complete Guide 2026 and the MCP roadmap breakdown.

Orca is a development environment (ADE) that runs multiple coding agents like Claude Code and Codex in parallel git worktrees — it has no heartbeats, budgets, or approval gates like Paperclip. See the full Orca guide for a tool focused on parallel review of a single coding task. CrewAI is a Python framework for coding the coordination and role logic between agents; it doesn't ship a human approval gate or org-chart UI the way Paperclip does out of the box. OpenClaw is an autonomous AI agent itself — in the README's own metaphor, closer to "the employee Paperclip hires." See What Is OpenClaw? for its basics.

- Paperclip doesn't run agents itself — contracts, API keys, and authentication for Claude Code, Codex, OpenClaw, and any other agent are your own responsibility (bring your own agent)
- The README itself says "if you have one agent, you probably don't need Paperclip. If you have twenty, you definitely do" — it can be overkill for light, single-agent use
- It explicitly states it is not a workflow builder, prompt manager, or code review tool, so drag-and-drop pipelines and the review process itself are out of scope
- Company Portability lets you export/import an entire organization (agents, skills, projects, routines, issues), but check its secret-scrubbing and collision-handling behavior before relying on it
- Runaway-cost protections exist (budget auto-stop, atomic checkout), but final billing should always be confirmed in each agent provider's own console

It suits individuals or teams already running several AI agents who have lost track of what each one is doing, can't predict cost, or want a real approval process. If you're only using a single coding agent, a lighter tool like Orca above may be a better fit.

Rather than making a single agent smarter, Paperclip is a control plane for running multiple agents as an organization — goal-setting, delegation, execution, budgets, and human approval gates all live in one dashboard, aimed at the chaos of manually managing twenty-plus agent sessions.

That said, contracts and authentication for the agents themselves remain your responsibility, and it can be overkill for single-agent use. Trying a small organization with the test-drive command first, then tightening budget and approval rules as you scale, looks like the reasonable path to a full rollout.

- https://github.com/paperclipai/paperclip
- https://paperclip.ing
- https://docs.paperclip.ing
- https://github.com/paperclipai/paperclip/releases/tag/v2026.916.1
- https://github.com/openclaw/openclaw/blob/main/LICENSE
- https://github.com/trending?since=weekly