Paperclip review: the open-source AI agent runtime, tested

What Paperclip actually is

I ship agent and integration code for a living, so let me describe Paperclip the way I'd describe it to a teammate, not the way its landing page does.

Paperclip is a Node.js server and React UI that orchestrates a team of AI agents to run a business. It's MIT-licensed, sits at roughly 74k GitHub stars, and the whole pitch is governance. You give your agents an org chart, assign monthly budgets that auto-pause at 100%, set goals that trace back to a mission, and route work through a ticket system where "every conversation is traced, every decision explained." Agents wake up on scheduled "heartbeats" and delegate work down the chart.

The clever part is that it's agent-agnostic. Claude Code, Codex, Cursor, OpenClaw, and Hermes all plug in as the actual workers, while Paperclip is the layer that hires, budgets, and audits them. Its own docs are refreshingly honest about the boundary: it's explicitly not a chatbot, not a single-agent tool, and not a support product.

That one line is the most important thing in this whole review, so I put the gap in a diagram.

Paperclip governs the agents. It does not include the thing on the other end that a support team actually needs, the helpdesk where tickets get resolved.

Should you even run Paperclip?

Before the deep dive, answer two quick questions. This is a shortcut, not gospel, but it'll tell you whether the rest of this review is for you or whether you should jump to the alternatives.

What Paperclip gets right

Credit where it's due, because Paperclip earns it. If you're building an agentic AI operation, these are real strengths.

• Governance is the whole point, and it's done well. Budgets that auto-pause at 100%, goals that trace to a mission, and a full audit trail on every decision. For anyone nervous about a fleet of agents running unsupervised, that's the right instinct baked into the architecture.
• Truly agent-agnostic. Because it treats the underlying agent as swappable, you're not locked to one model or vendor. Swap Claude Code for OpenClaw without rewriting your org chart.
• Self-hosted and MIT-licensed. Your data and your agents stay on your infrastructure. For teams with hard data-residency rules, that ownership is worth a lot.
• The mental model is fresh. "Run agents like a company" is a clearer framing than most orchestration tools manage, and the heartbeat-plus-org-chart design makes long-running, delegated work legible in a way raw scripts don't.

As a piece of open-source engineering, it's a pleasure to read. I mean that.

Where Paperclip falls short

Now the honest part, because a review that only praises isn't a review. Most of these aren't bugs, they're consequences of what Paperclip chose not to be.

• No helpdesk on the other end. There's no native Zendesk, Freshdesk, or Gorgias connector. Paperclip manages agents; it doesn't sit inside your ticketing tool. For support, that's the whole ballgame missing.
• No agent of its own. Paperclip is the control plane above the agents, so you still bring, configure, and pay for the actual AI agent and its model. It's an org chart with no employees until you hire them.
• No way to test on your data. There's no simulation against your past tickets. Here's where the years my team has spent putting agents on live queues matter: we've watched a confident-sounding bot quietly give a wrong answer, which is exactly why eesel now simulates every rollout against real history first. Paperclip's audit log tells you what happened; it can't tell you what will happen before a customer is affected.
• You run the stack. Node 20+, pnpm, a Postgres database, and a server, all yours to install, patch, scale, and babysit.
• Cost is hard to forecast. Every heartbeat is an LLM call. Budgets pause work at a ceiling, but they don't lower the per-call price, so a busy org chart can burn tokens faster than you'd expect.

None of this is a knock on the project. It's the difference between a tool that helps you run a company of agents and a tool that is the support agent.

Paperclip pricing: free, but not free

Let me be precise, because "free and open source" hides the real number.

The Paperclip runtime is free under the MIT license, with no account and no seat fees. What you actually pay for is everything around it: the LLM API calls each agent makes (the biggest and least predictable line), the server and Postgres database to host it, and the engineering time to keep the whole thing running. For a support use case, the number that matters is cost per resolved ticket, and a self-hosted agent fleet makes that hard to pin down in advance.

Contrast that with usage-based pricing where you pay per resolution and infrastructure is someone else's problem. Neither is "cheaper" in the abstract, but one gives you a forecastable per-ticket number and the other gives you a variable infra bill. Know which one you're signing up for.

Who should run Paperclip, and who shouldn't

Strip away the hype and it comes down to fit.

Run Paperclip if you're an engineering-heavy team building an autonomous agent operation across many agents, you value owning the stack, and governance (budgets, audit trails, org charts) is a real requirement rather than a nice-to-have. For that job, it's one of the best open-source options going, and my full Paperclip alternatives piece covers the CrewAI and LangGraph crowd you'd compare it against.

Skip Paperclip if your actual goal is to answer customer tickets. You'd spend weeks bolting on a helpdesk connector, an agent, and safety testing, only to reinvent a customer service AI that already exists. That's not what Paperclip is for, and forcing it into that shape is the fastest way to be disappointed by an otherwise excellent tool.

The support-shaped hole, and how I'd fill it

If you read this far because you wanted an AI on your support queue, here's the direct answer: you don't need an org chart, you need a teammate that resolves tickets.

eesel is the opposite bet from Paperclip. Instead of building a company of agents, you plug one AI teammate into the helpdesk you already run and let it answer tickets. It connects natively to Zendesk, Freshdesk, Gorgias, and Help Scout in minutes, trains on your past tickets and docs automatically, and, crucially, lets you simulate on your real history so you can see its resolution rate before it replies to a live customer. Paperclip audits what happened; eesel shows you what will happen.

"In the first month, eesel is resolving 73% of our tier-1 requests. We saw results quickly during our 7-day trial."

Kim Simpson, Gridwise (G2 review)

Try eesel free, or book a demo to see it run on your own tickets.

My verdict

Paperclip is a wonderful way to run a company of agents, and as an open-source project it's a genuine standout: thoughtful governance, real ownership, and a mental model that's clearer than most. If that's the job you have, it earns the stars it's collected.

It's just not a support agent, and no amount of org-chart governance changes that. Pick the tool that matches the job you actually have: Paperclip to orchestrate a fleet, a helpdesk-native AI to resolve tickets.