Paperclip AI Agent Orchestration: Run Your Agents Like a Company
Prefer to watch? Watch on YouTube · Subscribe for new builds · Read the transcript
In this video I break down Paperclip, an open-source GitHub repo that puts a management layer on top of your AI agents, so they run like a company and not a pile of separate bots. It's for anyone running more than one agent who wants org charts, budgets, and oversight without building all of it by hand. I recorded this in March 2026, when the repo had over 31,000 stars. Tools like this change fast, so check the current docs before you set anything up.
What I cover in this video
- 0:00 My current setup: 13 AI agents on a Mac Mini that I built by hand over 45 days
- 1:31 What Paperclip is: "If OpenClaw is an employee, Paperclip is the company"
- 2:16 Org charts for AI agents, with real reporting lines and delegation
- 4:33 Budget enforcement and multi-company support
- 6:04 The dashboard, agent-agnostic design, and governance
- 7:36 Persistent state, Clipmart templates, and the one-command install
- 11:26 My honest take: it's early, and agent drift is a real issue
- 12:12 A real scenario: setting up a content business with Paperclip
The problem Paperclip solves
Right now I run 13 agents on a Mac Mini. Some scout trending topics every morning at 5:00 a.m., some write video scripts, some make thumbnails, some write blog posts and draft newsletters, and one runs my whole X account. I built all of it from scratch with cron jobs, memory files, a Neon database, a custom dashboard I call Mission Control, Supermemory for sharing context between agents, and an org chart I set up myself. It works, but it took 45 days, and honestly it's held together with duct tape.
Most of us have been focused on making single agents better, like making Claude Code smarter or OpenClaw more capable. The harder question is how you manage a team of them. How do you get 20 agents working toward one goal? How do you stop one agent from burning through $300 in API credits while another one sits there doing nothing? Paperclip is built to answer that. It's the management layer.
The 10 features I walk through
- Org charts. You can set up a CEO agent, a CTO agent, and engineering, design, and marketing agents, with real reporting lines. Engineering reports to the CTO, the CTO reports to the CEO, and tasks move through that structure. I already do this by hand. My Ghostface agent scouts intel, my RZA agent writes script outlines, and my U-God agent makes thumbnails. Yes, they're named after the Wu-Tang Clan.
- Goal-aligned execution. Every task traces back to the company mission. Context flows from the task up through the project to the company goal, so agents know why they're doing the work and not just what it is.
- Budget enforcement. This one is personal for me. I've had agents run in loops and eat into my usage limits on a $200/month plan before I noticed. Paperclip gives each agent a monthly budget, and the agent stops when it hits the limit. Enforcement is atomic, so there's no double spending and no race conditions.
- Multi-company support. One deployment can run many companies with complete data isolation. Think of a content agency, a SaaS product, and a consulting practice, each with its own agents, goals, budget, and audit trail, all managed from one control plane.
- Dashboard. You can see agents, costs, tasks, and conversations in one UI, and you can manage it from your phone. You can check what your CTO agent is doing and approve a task while you're out.
- Agent agnostic. It works with OpenClaw, Claude Code, Codex, Cursor, and others. The GitHub page puts it this way: "If you can receive a heartbeat, it's hired." You choose the best agent for each job.
- Governance. You're the board. You approve hires, override strategy, and can pause or terminate any agent at any time. A full audit log records every conversation, decision, and tool call, and you can roll back if something goes wrong.
- Persistent state. Agents keep the same task context across heartbeats, so they don't start from scratch every time they wake up. I had to build this myself with memory files and Supermemory.
- Clipmart. This was still marked "coming soon" when I recorded. It's basically an app store for whole companies: pre-built templates with org structures, agent configs, and skills that you import into your Paperclip instance. It's the same idea as my Content Machine product, which is a pre-built team of 10 agents.
- One-command install. One command starts the API server and creates an embedded PostgreSQL database for you.
How to install Paperclip
This is the install command from the repo description:
npx paperclipai onboard --yes
You don't need Docker or any complicated config. All you need is Node.js 20 and pnpm. It's open source, MIT licensed, self-hosted, and doesn't require an account.
My honest take
Paperclip is early. When I recorded this, people on Reddit were rating it around 6 or 7 out of 10. Some people have run into agent drift, where agents make decisions that don't match what you actually wanted. I've seen that with my own agents too, and I think it's the biggest challenge in getting OpenClaw agents set up right. There's still a lot of interest, though. There's already a dedicated subreddit, and one person set up a $20 AI company and let it run for two hours.
To me this feels like the early days of OpenClaw. It's not perfect, but it's where things are going. At GTC, Jensen Huang said every company needs an OpenClaw strategy and compared it to Linux and HTML. If that turns out to be true, the management layer for AI agents will matter just as much.
Here's a concrete example. If you wanted to build a content business today, you'd have to set up writer, editor, SEO, social, and research agents yourself, then work out who reports to whom, how tasks get handed off, and how budgets work. That's weeks of setup. With Paperclip, and eventually Clipmart, the plan is that you download a content company template with a CEO agent running strategy, an editor agent reviewing, writers producing, and a social agent handling distribution. Then you set your goal and budget and start it.
Key takeaways
- Paperclip doesn't replace OpenClaw, Claude Code, or Codex. It runs on top of them and manages them like a company.
- The core features are org charts, goal alignment, per-agent budgets, governance with audit logs, and persistent state.
- Per-agent monthly budgets that stop agents automatically can prevent runaway token costs.
- You install it with
npx paperclipai onboard --yes, and it needs Node.js 20 and pnpm. - It's early. Expect rough edges like agent drift, and keep a close eye on what your agents decide.
FAQ
What is Paperclip AI?
Paperclip is an open-source, self-hosted Node.js platform that organizes AI agents into a company structure, with org charts, budgets, governance, and goal alignment. Its tagline is: if OpenClaw is an employee, Paperclip is the company.
How do you install Paperclip?
Run npx paperclipai onboard --yes. It starts the API server and creates an embedded PostgreSQL database automatically. You need Node.js 20 and pnpm, and no Docker setup.
Does Paperclip work with Claude Code and OpenClaw?
Yes. Paperclip is agent agnostic and works with OpenClaw, Claude Code, Codex, Cursor, and others. The GitHub page says, "If you can receive a heartbeat, it's hired."
Is Paperclip ready for production use?
When I recorded this in March 2026 it was still early. Reddit users were rating it about 6 or 7 out of 10, and some reported agent drift. The architecture is solid and it's MIT licensed, but expect rough edges.
Transcript
Read the full transcript
0:00This GitHub repo absolutely blew my mind and by the end of this video, it's going to change your mind on how you think about open claw and AI agents forever. Because what I'm going to tell you is just not another AI tool. This is the missing piece that everyone that's running AI agents has been looking for and I mean everyone. I do have it linked down below. You got to check it out. So let me set the stage for you. Right now, I'm running 13 AI agents on a Mac Mini and I've got agents that scout the internet. I got them scouting trending topics every morning at 5:00 a.m. I have agents that write video scripts, agents that create thumbnails, agents that write blog posts, agents that draft newsletters, and also one agent that manages my entire X account. And you know what? I built all of this by
0:46scratch over the last 45 days. You know, cron jobs, the execution agent, memory files, got everything on Neon database, custom dashboard that I call Mission Control, super memory for sharing context between agents, org charts that I literally decide myself. So every agent knows its role and I wish I had known about this when I first started. It works. It did take 45 days to build. I'm very proud of it. I learned a lot along the way. But what if I told you there's now a single open source tool that does all that for you, one repo, one install command, and it's already got 31,000 stars on GitHub. Ladies and gentlemen, meet the repo called Paperclip. I'm sure you heard about it, but let me break down what it actually
1:31does and see if it could be beneficial for you. So what is it exactly? The tagline on their GitHub repo says it best. If open claw is an employee, Paperclip is the company. And that right there, the single line, that's what made me stop scrolling and actually check it out. Because here's the thing. We've all been so focused on making individual agents better, making Claude code smarter, making open claw more capable, getting Codex to write better code, and that's great. But nobody was solving the real problem. How do you manage a team of these things? How do you coordinate 20 agents working toward the same goal without losing your mind? How do you stop one agent from blowing through $300 in API credits while another one sits idle doing nothing? That's what Paperclip solves. It's the management
2:16layer. So I'm going to break down the features because it's pretty stacked. The first one is org charts for AI agents. I'm not even joking. You could set up a CEO agent, a CTO agent, engineering agents, design agents, and marketing agents. Real hierarchies, real reporting lines, and the delegation flows up and down the org chart just like a real company. Your engineering agent reports to your CTO, your CTO reports to your CEO, and every task flows through that structure. Now when I first saw this, I literally sat back in my chair because this is exactly what I built manually. My Ghostface agent scouts intel. My Rizza agent writes script outlines. My U-God agent makes the thumbnails. And yes, they are named after the Wu-Tang Clan. That's the org chart situation there. I just never had
3:02the tool that made it official. But Paperclip does make it more official. So that's amazing. You got to check it out down below, linked. Real quick, before we keep going, if you're watching this and you want to actually build with some of these tools, not just watch videos about them, then you're going to want to check out our community down below, Shipping School. We have a full Claude code course, a full open claw course, and four live bootcamps every single week where we actually help you get set up from scratch. Like actually set this thing up, not just watch a tutorial and figure it out by yourself. And we also provide one-on-one coaching, so you could book a call with me, we could share screens, and I can help you get Claude code or open claw running on your machine. That's it, no fluff. I built this community because watching
3:47YouTube's only gets you so far. We launched it just 3 days ago and we have over 55 members. You need people around you who are actually building, people who hold you accountable, and coaches who can help you when you get stuck. I'll put the link in the description down below. Get in now before the price goes up. The goal-aligned execution is the second thing. This one's pretty big. Every single task in Paperclip traces back to the company mission. So like you set a goal, you know, for example, build number one is AI note-taking app to $1 million monthly recurring revenue, make no mistakes. And then every agent, every task, every decision that happens in your company connects back to that goal. That's pretty cool. Your agents aren't just doing random work. They know what they're doing and why they're doing it. Context flows from the task up through
4:33the project, up through the company goals. Your agents always know the why, not just the what. The third thing, this one's huge, is budget enforcement. And can I get an amen for this? Because this one thing right here, this is personal for me. I have agents that blow through credit usage like it was nothing. And if you knew you've been using Anthropic models these last few days, you know what I mean. Just burning tokens and you're like, wait, I'm on the $200 a month plan. How am I hitting my usage limit already? But these agents just run in loops at times and by the time I caught it, you're already hitting out your max plan. You're already hitting usage limits. But Paperclip gives you monthly budgets per agent. And when you hit that limit, they stop. So no runaway costs. And it's atomic, too, you know,
5:18which it means there's no double spending, no race conditions, no agents sneaking past their budget while another one checks out the same task. That's the kind of detail that tells you real engineers built this thing. The fourth one is multi-company support. This is where it gets really interesting. One deployment, many companies, complete data isolation. So imagine you've got three different AI businesses, right? A content agency, a SaaS product, and a consulting practice. Each one has its own team of agents, its own goals, its own budget, all running on the same Paperclip instance. Separate data, separate audit trails, and one control plane for your whole portfolio. Now if you're thinking that sounds like what a holding company does for real businesses, yeah, that's pretty much
6:04exactly what it does, but for AI companies. The fifth thing is the dashboard. You can monitor your agents, all your costs, all your tasks, all your conversations from one single UI. And you can manage it from your phone. So you could literally be at the grocery store, at Taco Bell, at McDonald's, wherever you are, pull out your phone, check what your CTO agent is working on, approve a task, and put the phone back in your pocket. That's the world we're living in and I love it. The sixth thing is agent agnostic. So this is the key. Paperclip works with open claw, Claude code, Codex, Cursor, whatever you want. Their exact words on the GitHub page are, "If you can receive a heartbeat, it's hired." I love that because it means you're not locked into one provider. You pick the best agent for each job. Maybe your coding agent runs Claude code. Maybe your research agent
6:51uses open claw. Maybe your marketing agents use something completely different. Paperclip doesn't care. It manages the company. Your agents bring their own skills. That's amazing. The seventh thing is governance. You're the board. You approve the hires. You override the strategy and you can pause or terminate any agent at your any time. There's a full audit log so you can see every conversation, every decision, every tool call that any agent made. And if something goes wrong, you could always roll it back. This isn't some black box where you hope your agents are doing the right thing. You have complete visibility and complete control. The eighth thing is persistent state. So your agents resume the same task context across the heartbeats. They don't restart from scratch every time they
7:36wake up. Think about how important that is. If your engineering agent was like halfway through building a feature and then the session ends, it picks up right where it left off with the same context, same progress, no wasted tokens, re-explaining what the project is. This is something I had to build myself with memory files and super memory. Paperclip just does it out of the box. Number nine is Clipmart. Now this is one that is coming soon, but I'm ready and really excited for it. Clipmart is basically an app store for entire companies. You go to Clipmart, you browse pre-built company templates, full org structures, agent configurations, skills, basically everything. And you download it and you import the entire company into your Paperclip instance and in seconds, you
8:22got it. One click. That does that sound familiar? I mean, I it sounds familiar to me. It should, really, because that's basically what my content machine product is on Claude Mart doing right now. It's basically a pre-built team of 10 agents with roles, schedules, feedback loops that you download and you run. And we made it so easy where you just download the file, give it to your agents, and it implements the systems and the pipeline. But Clipmart is taking that same concept and making it the standard for the whole ecosystem, which is pretty freaking cool. The 10th thing in this whole GitHub repo is one command install. I'm just asking you, like, are you ready for this? This is insane. NPX Paperclip AI onboard dash dash, and that's it. Crazy. One command that starts the API server, creates an
9:07embedded PostgreSQL database automatically, and you're up and running. No Docker setup, no complicated configs, Node.js 20, and PNPM, and that's all you need. So I'm going to zoom out for a second because I want you to understand why this matters so much. Right now, most people think of AI agents as chatbots doing one thing. We're way past that. You can open Claude, you ask it a question, it gives you an answer, or you set up an open claw agent to do the same task. That's cool, great. That's one employee. But what happens when you have five employees, 10, 20, 100, whatever? How do you coordinate them? How do you make sure they're all working toward the same goal? How do you track the costs? How do you handle the delegation? How do you audit what they did? You need a company structure. You need management and then that's Paperclip. They built it all out.
9:53Here's where it connects to what I've been building. And I've been living this problem for weeks, building my own version of agent orchestration with backs with bash scripts and cron jobs and database tables. It works, but it took a long time to get to this point. It's a duct tape basically. It's all built together custom for me. You know, this what I'm talking about compared to paper clip. You know, it's purpose-built tool with 31,000 stars and a growing community behind it. If you're struggling to keep up with content, what I'm about to save you about 40 days worth of work. [music] I built something called content machine. It's 10 AI agents that run on the open claw orchestration and they handle everything. Scripts, thumbnails, X posts, blogs, outreach, clips,
10:38newsletters, all of it. So, I went from 1,000 subscribers to 4,000 subscribers on YouTube in 7 days using this exact system. Every single morning, I wake up [music] and the content's already done. I spend maybe 15-20 minutes reviewing and approving them and I move on with my day. It works for any niche, fitness, finance, [music] real estate, marketing, whatever you are building and it is 100% completely customizable to your use case. So, you get the mission control dashboard, all of the cron jobs, everything I've built over the last 40 days helping me gain more and more people to subscribe and join the community. So, you plug in your own thing and it molds it to you. It learns how you talk and it writes so it doesn't sound like AI slop. $97 one
11:26time, it's not a subscription. I'll put the link down below and you'll thank me later. I want to be honest with you because that's how we do things on this channel. Paper clip is early. People on Reddit are giving it a six or a seven out of 10 right now. Some folks have had issues with agent drift where agents make decisions that don't align with what you actually wanted. And I've seen that myself with my own agents. And it's the biggest challenge of getting open claw agents set up correct. But the Reddit threads are also buzzing about this. People are experimenting with it. There's already a dedicated subreddit. YouTubers are already starting to cover it. Someone literally set up a $20 AI company and let it run for 2 hours. And the architecture's pretty solid. It's open source, MIT license, self-hosted, and no account required. The way that I
12:12see it is this is like the early days of open claw. Is it perfect? Absolutely not. But it is the direction where this things are headed and and I love it. Because think about what Jensen Huang said last week at the G TC conference. He said every company needs an open claw strategy. He compared it to Linux, HTML. And if open claw is going to be as fundamental as Jensen says, then the management layer for open claw agents is going to be just as important. And right now, paper clip is leading the open source answer to that. So, I'm going to give you a real scenario. Say you want to build a content business. Right now, you have to set up a writer agent, an editor agent, an SEO agent, a social media agent, a research agent. You'd have to figure out how they talk to each other, who reports to who, how tasks get
12:58delegated, how budgets work. And that's weeks of setup. I know because I did it 45 days ago. With paper clip and eventually clipmart, you download a content company template. CEO agent manages the strategy. Editor agent reviews everything. Writers produce the content. Social agent does the distribution. And it's all pre-configured, all coordinated. You just set your goal and your budget and press go. That's the zero human company right there. Not one chatbot. An entire company running autonomously with AI in every role. Now, if this type of stuff excites you like it does me and if you like building with AI agents and if you're trying to turn your ideas into real products and real businesses, you should check out our community. I have a link down below. 160 people have joined
13:45in the last 15 days. We do six live bootcamps every single week with AI experts and myself. We also do a content call every Saturday to help you get your first customers. It's a community where builders like you and me are figuring it all out together. You know, real people building real things. Like I said, six live calls a week and I love everyone in there. We're growing. We're building in public. We're building things. We're making progress. That's all that matters. Our whole goal of the community is to help you take the next step in your AI journey. So, paper clip isn't just a replacement for open claw. It's not a replacement for Claude coder codex. It uses those tools. It's the layer above. If your AI agents are employees, paper clip is the company that manages them. The org chart, the
14:30budget, the goals, the governance. I've been building my own version of this for over 4-5 weeks now. 13 agents, manual orchestration, custom dashboards. And now there's an open source tool with 31,000 stars on GitHub for free that does it all. Would I have saved myself weeks of work if this existed when I started? Yeah, but that's okay. I learned a lot. Am I going to be watching this project very closely going forward? You believe it. I love it. I got it bookmarked. And it's in our community too on how to install it and implement it. The zero human company isn't science fiction anymore. It's a GitHub repo with 31,000 stars and one line install command. Go check it out. You got to look at it. Start the repo, try it out. Comment down below what are your thoughts or maybe you've been using it already. And please subscribe to the
15:15channel because we release about four to six videos every single day. Have a blessed day, guys. Bye-bye.
Ready to start building with AI?
Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.
Join Shipping SkoolKeep reading
Claude Code Agent Teams: The Observer Pattern for Multi-Agent Coding
The observer pattern adds one read-only agent to your Claude Code team. It watches every parallel worker through file changes, status files, and logs, then writes a shared summary of what's done, what's in progress, and what's conflicting. Nobody has to dig through each agent's work to figure out where things stand.
ClawTeam Explained: Multi-Agent Swarms for OpenClaw and Claude Code
ClawTeam is an open-source, MIT-licensed repo from HKUDS. It turns CLI agents like Claude Code, Codex, OpenClaw, and Cursor into a team: a leader agent splits up the goal, then sub-agents work in parallel in their own Git worktrees and talk to each other through an inbox system.
Claude Code Channels vs OpenClaw: Is It Really an OpenClaw Killer?
Claude Code Channels lets you message one Claude Code session from Telegram or Discord, and it takes about 5 minutes to set up. It's not an OpenClaw killer, though. OpenClaw supports 15+ platforms, any model, scheduled multi-agent pipelines and custom memory. Channels wins on security and ease of setup.
