Pancake vs Paperclip AI: Setup, Cost, and Who Each One Is For

Paperclip AI is an open-source app for running a team of AI agents as if it were a company. It launched in March 2026 and passed 80,000 GitHub stars by September. Before you install it, three practical questions matter. How hard is it to set up? What does it cost to run? Is it the right tool for your job?

This page answers those questions from Paperclip's own README, website, and release notes as of September 2026. Then it covers where Pancake fits, for readers whose real goal is customers rather than an agent org chart. If you first want to know whether you need an agent control plane at all, read Pancake vs Paperclips.

TL;DR: Paperclip is free, self-hosted infrastructure for coordinating agents you bring yourself. It suits technical operators who run many agents across coding, research, and content. Pancake is an AI GTM team whose agents find your buyers and run your outreach. Pick Paperclip to build and govern an AI organization. Pick Pancake if the job is pipeline and you would rather not build it.

---

What Paperclip AI is

Paperclip is a Node.js server with a React dashboard, released under the MIT license by Paperclip Labs. The code lives in its GitHub repository. The README sums up the idea in one line: "If OpenClaw is an employee, Paperclip is the company."

You run it like a small company. You write a goal, such as a revenue target for a product. You hire a CEO agent, which proposes a strategy and asks to hire more agents: a CTO, engineers, a marketer. You sit on the board and approve or reject each hire. Agents then wake on a schedule, which Paperclip calls a heartbeat. They check their tickets, do the work, and report up the org chart.

Paperclip ships no agents of its own. You bring them: Claude Code, Codex, Cursor, Gemini, Hermes, OpenClaw, OpenCode, Pi, a Bash script, or any HTTP endpoint that can receive a heartbeat. Paperclip manages the organization around them. That means tickets, goals, per-agent budgets, approvals, and an audit log of every tool call. One install can run several companies with their data kept apart.

The README also lists what Paperclip is not. It is not a chatbot, an agent framework, or a drag-and-drop workflow builder. That line matters when you evaluate it. Paperclip manages who works on what and how much each agent may spend. The agents decide how the work gets done, with the prompts, models, and tools you give them.

From "zero-human company" to managing agents at work

Early copies of the repository described Paperclip as open-source orchestration for "zero-human companies." Much of the coverage still uses that phrase. By September 2026 the tagline had moved to managing AI agents for work. The website now pitches it to engineering, support, sales and marketing, operations, and finance teams.

The product still puts a human in charge. You are the board. Agents can't hire other agents without your approval, the CEO agent can't execute a strategy you haven't reviewed, and public work can wait at an approval gate. A hands-on review published in April 2026 shows why those gates matter. The reviewer found the cold email drafts decent. The same run produced marketing statistics with no source and a landing page with a broken layout.

What setting up Paperclip involves

The install is quick if you are comfortable in a terminal. You need Node.js 24.11 or newer. One command, npx paperclipai onboard --yes, starts onboarding. A local instance runs with an embedded PostgreSQL database and local file storage, so there is no database to set up on day one. For production, you point Paperclip at your own Postgres and deploy it on a server you choose.

The configuration takes longer than the install. Plan for five steps:

1. Choose your agent runtimes. Each agent needs an adapter and credentials: an API key, or an existing subscription to Claude Code, Codex, or another provider.
2. Write the roles. Each agent gets a title, a job description, and instructions. Paperclip reads AGENTS.md files and has a skills manager for skills shared across the org.
3. Connect the tools. Paperclip comes with no business tools. Your agents use what you connect: MCP servers, extensions, webhooks, GitHub access and, since the September 16 release, experimental agent email addresses through AgentMail.
4. Set budgets. Every agent gets a monthly spending cap. Paperclip warns you at 80% and pauses the agent at 100%.
5. Review the CEO's plan. The CEO agent proposes a strategy and the hires it wants. You approve it before the work starts.

Two more details are worth knowing before you install. Anonymous usage telemetry is on by default, and you can switch it off with the DO_NOT_TRACK=1 environment variable. The main button on the Paperclip homepage is a waitlist for a hosted version. Today, the version you can run is the self-hosted one.

What Paperclip AI costs to run

The software is free. As of September 2026, Paperclip publishes no paid plan, and the local version needs no account. The costs sit in three other places.

Model usage. Every heartbeat is a run on your model provider. It is billed to your API key or counted against your subscription. More agents and shorter heartbeat intervals mean more runs. Budgets stop an agent at its cap. They don't lower the price of each run.

Hosting. Your laptop costs nothing extra. If you want agents working while the laptop is closed, you need a server.

Your time. Someone writes the roles, connects the tools, reads the tickets, and approves the work. For a solo founder, it can be the largest line.

Paperclip is good at showing the first cost. It breaks spend down by agent, task, project, goal, and model. You can see which role costs the most and move mechanical work to a cheaper model.

What Paperclip does for sales and marketing

Paperclip has a sales and marketing page, and its scope is research and content. Research agents run recurring sweeps of competitor releases, pricing changes, and industry news, then deliver briefs with citations. Content agents draft blog posts, comparison pages, release notes, and social posts against your style guide. Every public artifact stops at an approval gate with the final version attached. Paperclip says its own blog works this way: an agent drafts each post and a human approves it.

The page does not describe lead generation. Paperclip itself ships no lead database, no buying-signal feed, and no outreach sequence. An agent can prospect only through tools you connect. You also write the logic: who counts as a good lead, what the first message says, when to follow up, and when to stop.

A technical team can build that loop on Paperclip. If you want the loop built already, a GTM product is the shorter path.

Where Pancake fits

Pancake is an AI GTM team for founders and small B2B companies. It does the part Paperclip leaves you to build: finding people who show buying signals and reaching out to them.

Setup starts with your website. Pancake reads it and builds what it calls the GTM Brain, one shared memory of your positioning, ideal customers, offers, proof, and objections. It keeps learning from what works.

Pancake then watches six kinds of signals:

• Keyword: people posting about a topic you choose, such as "ticket backlog".
• Competitor: people engaging with a rival's posts.
• Influencer: people engaging with posts from a voice your buyers follow.
• Own brand: people engaging with your own posts.
• Hiring: companies with open roles that match your buyer.
• Stack: companies whose job posts name a tool, such as one you replace.

New leads arrive each morning, and each lead says which signal picked it. They show up in Slack too, and you approve them there or in the app. Outreach runs from your own account on the professional network. The sequence is a profile visit, a like on a recent post, an invite with no note, and up to three messages. The first message makes no pitch. It asks one light question tied to the signal.

Pancake also writes articles built to show up in Google and AI answers. Approvals on leads and articles. You stay in control.

If your team already works in Claude, ChatGPT or Codex, Pancake meets you there. Connect its MCP server and you can check the day's leads, tune a signal or start outreach without leaving the chat.

Your coding and back-office agents stay in Paperclip, while Pancake takes the go-to-market work. The price is $99/month flat, with every agent included, no tiers, and no seats. The 3-day free trial needs a credit card to start, and you can cancel anytime.

Paperclip AI vs Pancake at a glance

Who should choose which

Choose Paperclip AI if

• You already run several agents, such as Claude Code, Codex, or OpenClaw, and lose track of which one is doing what.
• You want one place to govern agents across engineering, research, content, and operations.
• You want your data, prompts, and API keys to stay on your own servers.
• You like building the system and are comfortable writing roles and wiring tools.

Choose Pancake if

• Your bottleneck is pipeline, not agent coordination.
• You sell B2B and your buyers leave public traces: posts on a topic, engagement with a competitor, open roles, the tools they use.
• You want outreach running from your own account without writing prompts or connecting APIs.
• You want a fixed monthly price with nothing variable.

Running both

The overlap is small. Both can produce articles, and that is about it. A technical founder could run Paperclip for coding and research agents and Pancake for finding buyers, each on its own.

Verdict

Paperclip AI earns its stars. It takes the mess of many agents in many terminals and gives it an org chart, a ticket queue, budgets that hold, and a board that signs off. If the organization itself is what you want to build, start there. Budget real hours for setup and keep the approval gates on.

If what you want from AI agents is customers, the org chart is overhead. Pancake skips it and starts from your website. It turns buying signals into leads you approve and runs the outreach from your own account.