Paperclip AI Tutorial: How to Build a Zero-Human Company

Paperclip lets you build a company run entirely by AI agents. A CEO that hires its own team, delegates tasks, and actually gets work done. In this video, I'll walk you through the full setup from zero to a working AI company that sends real output to your inbox. I'll show you what Paperclip is, how to get it running on a server, how to connect it to external tools like search and email, and what to expect when your agents start working. By the end, you'll have your own Paperclip company live on a server with AI agents that organize themselves and produce something real. So first, what exactly is Paperclip? Paperclip is a tool that lets you run a business or a team inside a business almost entirely with AI. Instead of using one AI chatbot to do one job at a time, Paperclip lets you set up multiple AI agents, give them each a role, and have them work together on tasks without you telling them what to do next. The best way to picture it is this. If an AI agent is an employee, Paperclip is the company. The agent is one worker, Paperclip is the whole office. The org chart, the goals, the budget, the rules. You set the direction and the AI handles the work. Here's how it works in plain language. You create a company inside Paperclip and give it a goal like produce a daily tech news digest. You hire a CEO agent and that CEO reads the goal and starts making decisions. It hires other agents like a CTO to handle technical work, a CMO for content. It delegates tasks and the whole team starts working. Each agent wakes up on a schedule, checks if it has work to do, does it, and then goes back to sleep. Your company runs around the clock without you managing every single step. Paperclip launched in March 2026 and became one of the fastest growing open source projects in AI in just a few weeks. It's free, it's open source, and you can run it on your own computer or server. It's still early software actively being developed with new features shipping regularly, so it's a great time to get in and start experimenting. We're running it on a virtual private server today because Paperclip is designed to run around the clock. You can run this on your personal computer, but that means it stops the moment you close your laptop. A server keeps it going 24/7, which is the whole point. Your AI company works while you sleep. And if you want to get fancy later, the same server can also host a website that your agents manage. We're using Hostinger for our server because they offer a one-click deployment template that includes a reverse proxy to make your connection safer over the open internet. I have a link that gets you an additional 10% off on top of any promotions they're already running. To get started, use the link on screen or click the first link in the description below. Using that link will take you to this page here and I want to point out that Hostinger has a 30-day money-back guarantee, so you can follow along with this process without committing. I recommend switching from the KVM 2 plan to the KVM 1 plan. KVM 1 gives you enough to run Paperclip and a few services on the same server. Once you've changed that, go ahead and click deploy. That'll take you to this cart page where you can see that the Benex Media 10% extra discount has already been applied to your order. The first thing you need to do is change the period for your order. You can choose 1 month, 12 months, or 24 months. You need to choose at least 12 months to take advantage of the coupon. If you choose 1 month, it will be removed. So let's do 12 months and that gets you in for the lowest upfront cost, plus you get a free domain for the first year with your order, which is cool if you want to spin up a website on your server later. Optionally, you can add auto backups and you can choose a server location. I recommend choosing the server closest to you or whatever has the lowest latency. Once you've got your settings on the cart page configured, go ahead and click continue. Next, register for an account. You can use Google, GitHub, or your email address. Personally, I'll use my email address. Then on the next page, enter your billing address and your payment information to complete your order. You'll get a confirmation screen and you can click the button or wait for the countdown to finish. Hostinger has now set up a server for us and now it's time to configure Paperclip itself. This page might look a little bit different by the time you're watching this video because both Paperclip and Hostinger get updated regularly. The first field is the admin name field. You can just leave this as is. Next, you have the admin email field. This is what you'll use to log into Paperclip. Go ahead and enter the email address you want to use. Next, you have the admin password field and Hostinger has automatically generated a secure password for you. To use this password, click the eye icon to see it, copy the password, and save it somewhere safe. Or you can always set your own password here. Next, you're prompted to connect an AI provider via an API key. Now, I'm going to show you later in this video how to connect to your OpenAI Codex subscription account. That way, instead of paying per use via the API, you're using your OpenAI subscription. For these other providers though, I highly recommend using the API key instead of trying to use your subscription cuz that could potentially violate the terms of service of these AI providers. To connect one of these models like Anthropic, grab the link from below the field, open that in a new tab, then follow the steps to add some credit to your account and get an API key, then plug in that API key here. But like I said, I'm going to show you in this video how to connect OpenAI Codex, so we'll just click deploy. On the next page, you might get a survey, but you can skip this if you want. Just scroll down and click skip. Next, I'll show you how to connect OpenAI Codex to your Paperclip. But first, we need to make sure your OpenAI settings are set correctly. In a new tab, open your OpenAI account by going to chat.openai.com, going to your account in the bottom left, and clicking settings. Then on the security page, scroll down to the bottom and make sure that enable device code authorization for Codex is turned on. If you skip this, the next step will give you an error and you'll have to start it over. After you've done this, you can close the chat.openai.com tab. Now, in the upper right of this Docker Manager page, click terminal. That opens the web terminal, which if you've never worked with one of these before can look a little bit scary, but essentially all you're doing is typing a few words to talk to the computer instead of clicking around with your mouse. I'll give you the exact commands you need here to connect your Codex account. First, click into the bottom, then type docker ps and hit enter or return on your keyboard. That will generate a response. The response has two different items. One of them is your Paperclip project and the other one is traffic. Under the names column, you'll see the container ID for Paperclip. Go ahead and double click that to highlight it and then command C or control C to copy it. Next, click back into the bottom, type docker exec -it exactly like I have it here on the screen. Put a space, paste in your container ID, use your right arrow key to deselect that container ID, add another space, then type /bin /bash. Then hit enter or return your keyboard. You're not going to get a response, but you will notice here that it changes from being root@server followed by a number to node@ followed by a string of numbers and letters. Now, instead of just talking with the server, we're talking with the specific Paperclip project. Next, type codex login --device-auth exactly like I have it here on screen. Then hit enter or return. This will get you a response that provides steps for how to log into your chat.openai.com So first, go ahead and copy this URL that it's providing, open a new tab, and paste that in. Go ahead and sign into your chat.openai.com On the next page, click continue. Then you'll be prompted to add a code. Go back to the terminal tab and copy the code from step two in the response. Return to the chat.openai.com and paste in your code, then click continue. You are now signed into Codex and you can close this tab. If you've done this correctly, you'll see in the terminal that it says you've successfully logged in. That's it. Codex is now connected to Paperclip. You can go ahead and close out this web terminal. Now that we're signed into OpenAI Codex on our server, let's open up Paperclip. In your Docker Manager page, make sure you're on the projects view, then find your Paperclip project, and under access, click open. Here to log in, enter the email and password you set on the Paperclip configuration page we saw earlier. We are now logged into Paperclip on our server and it's time to start building our company. We have four tabs we're going to go through here, creating the company details, your first agent details, your first agent's task, and then we'll launch. So first, let's name the company. I'll call mine Tech Co. Next, you need to give it a very high-level mission or goal. For me, as a simple company, what I want to do is build a email digest of the latest AI and tech news. So in this mission field, I'll enter find the biggest latest tech news and provide the world's best daily digest via email. The more high-level, goal and mission-oriented this description can be, the better. Once you've got that in there, click next. On the next page, we're going to set up your first agent or your CEO. By default, the agent name is set to CEO and you can leave this because even if you name it something else, the function of this agent will be your CEO. Next, you'll need to select your adapter type. The adapter connects Paperclip to your AI provider. If you set up Claude Code via API earlier, you can leave it set to Claude Code or if you set it to OpenAI Codex following along with the instructions, set it to Codex. If you use a different AI provider, you can click more agent adapter types to see the other options that are available and select the most relevant one. You can actually connect multiple different AI providers to your Paperclip organization, but each agent has to be set to only one AI provider. This adapter type is not setting the AI provider for the whole company, just for this particular agent. After you've selected your adapter, you can choose the specific model you want to use for your CEO. The default 5.3 Codex is fine for me, but by the time you're watching this video, there may be better models out there, so select the best one for you at the time you're going through these steps. Now, before we move on, we want to run a test to make sure our AI provider is actually connected correctly. Click test now and if everything is set up correctly, you'll get the green passed confirmation and you can click next. The next tab is the task tab where we're going to give your CEO its first thing to do. By default, it creates a task titled hire your first engineer and create a hiring plan and it has a description below. Feel free to tweak this slightly if you want. By default, the task is going to be to hire a founding engineer, write a hiring plan, and break the road map into concrete tasks and start delegating work. This first task gets things moving. You're telling the CEO to build a team and start working toward the mission. For most people, this is good enough for the first task. You can always add more tasks to your company later. So, go ahead and just click next. The final tab is the launch tab where you simply review everything you just set up. Go ahead and click create an open issue. Once you do that, you'll land on a page that shows you the task you just created. You can see the task title and description up top. And then if you scroll down, you can see a live run. This is the CEO actively working on this task. You can see as it thinks through the process and begins to take action. Your company is live and the CEO is already running. While this is working through the process, let's click on dashboard at the upper left. Here, if you scroll to the top, you can see all of the agents that are currently running. Right now, that's just our CEO. Below that, you have a high-level summary of everything on this page. There's one agent enabled, one task currently in progress, and in the third column, you can see your month spend. If you're on an API, this is more important than if you're on a subscription because every API call costs money. You can track that here. And here you see in the fourth column, the pending approvals just changed from zero to one. It's awaiting board review. Now, the philosophy behind Paperclip is that everything that happens within your Paperclip organization is like a company. All of the employees are agents. You, the human, are the board of directors. The CEO has put in a request that requires our approval. So, let's go check that out by clicking on the inbox on the left side. By default, all new hires or essentially new agents that get spun up need your approval. Here, the CEO wants to hire a CTO or chief technical officer. So, let's go ahead and approve this request. We're routed into an approval confirmation page. And if we click over to the dashboard on the left side again, we can see that the CEO is now thinking about this, following up on the task. And on the left sidebar, under agents, you can see that our CTO agent has already been hired. Right away, the new CTO gets to work as you can see here in the dashboard. And on the left side, under company, if you click org, you can see your company's current org chart with the CEO at top and the CTO reporting directly to the CEO. The CEO is responsible for the high-level mission of the company and delegating tasks to agents or employees below it. Just like a real-world organization, over time, this org chart will expand. The CEO will hire different C-suite agents, each with their own specialized departmental focus. And under those agents, you'll get even more specialized agent employees. Again, by default, all the requests to hire new agents will come into your inbox here. This is where all new messages to you, the board, and pending approvals will land. As time goes on, these toast notifications will pop up showing you what issues are being created. Let's take a look at the issues tab. That's under work on the left side. Think of issues like tasks in a project management software like Asana or Monday. You can look at all pending organizational tasks here in list view or with this upper right button, you can toggle from list to Kanban board view. Personally, I like looking at everything as a Kanban. On this issues tab, you can see all the individual tasks that are queued up in the to-do column and who they're assigned to as well as their priority. The CEO is currently working through that first task we gave it and the CTO has four tasks queued up in the to-do column. As these tasks get worked on, they'll move through to the right and land in the done column. They may also land in in review if a sub agent needs to have the agent that it's reporting to review its work first before it can move on. Tasks may also land in the blocked column if the agent that was assigned the issue or task can't finish the task due to some external blocker. For example, if it doesn't have permission or access to something that it needs. And then any issues or tasks that get canceled will land in this far right canceled column. Tasks that require your attention may also end up in blocked. So, pay attention to this column as time goes on. It looks like we have something new in our inbox. Let's go check it out. And it looks like the CEO wants to hire a CMO, chief marketing officer. Let's go ahead and approve this request. If you want to really control your costs, I definitely recommend keeping it set so that you have to manually approve all new hires. But if you're comfortable letting your organization act a little bit more autonomously, on the left sidebar, you can scroll down to settings. And then on the settings page, under hiring, you can toggle require board approval for new hires to off. This will make it so the CEO can go ahead and hire any agent it needs without having to ask you first. If we head back to our org chart, we can see that our organization is now expanding with the CEO orchestrating and delegating tasks with specific focuses to more specialized C-suite sub agents. Now that we've seen how the organization is structured, let's take a look at an agent detail page briefly. For example, let's click into the CEO. You can see your full list of agents on the left sidebar and select any agent you want from the list. Here, there are six key tabs. The dashboard where you can see all high-level details about this agent. You can see that there's currently a live run. The CEO is actively running. The next tab is instructions. Here are all the workspace files that tell the agent who it is, what its job is, what to do each time it wakes up, and what tools it has access to. The skills tab shows you what skills the agent has. Configuration shows more technical details like identity, what adapter or AI provider it's using, whether or not this particular agent has access to search the web, what model this agent is using, how often it should wake up and check for new tasks. This is called heartbeat, as well as permissions and API keys. Now, let's talk about the heartbeat really briefly. Essentially, the way this works is Paperclip will wake up the agent at a set interval to have it check for work. On each heartbeat, the agent reads the company mission, looks at its assigned tasks, does the work, and goes back to sleep. You can set this to whatever interval makes sense. It could be every hour, every 10 minutes, once a day. During the setup process, I recommend setting your CEO agent to a lower interval like 600 seconds, which is essentially every 10 minutes. That way it can work through some of the setup tasks quicker. But if you want to control costs and make it so that your organization doesn't burn through as many tokens as quickly, you can set this to a higher interval, too. Just make sure that once you've made any configuration changes, you click save. Back at the top, you can also track individual runs, see all the current runs as well as historical runs. And the final tab is budget where you can set the total monthly budget per agent. At any point, you can assign a specific agent a task by clicking assign task. And that'll open this new issue pop-up model. I'll go ahead and close this. You can also run the heartbeat manually at any time by clicking run heartbeat. And if the agent is currently running and you want it to stop, you can click pause. But we're just going to let our CEO keep running. While we've been sitting here talking through this, you can see that since we turned off approvals for the hiring process, a new agent has already been hired, the founding engineer. If we look at the org chart, you can see that now the CTO has a sub agent under it. It seems like based on the plan, this founding engineer is going to be doing a lot of legwork, building out the infrastructure to handle this company's needs. Right now, your agents can think and plan, but they can't reach the outside world yet. To make them actually useful, we need to connect external tools. We're going to give them two for this demo. A search API so they can research and an email API so they can deliver something real. You can follow these steps that I'm about to do for any external tool that has an API. For my first tool, I'm going to add Brave Search. So, we'll open a new tab and go to brave.com/search/api. On this page, we'll click get started. We'll fill out the fields to register a new account, then verify the email address by opening our inbox and clicking verify email on them email from Brave. Next, we get an account verification message and we can click the button to log in. On the next page, we can actually enter our password to log in. Then we need to grab the six-digit code from our inbox, copy that from the email, paste it in here, and submit. Next, we want to grab an API key. So, click API keys on the left side. And then here it says you need to go to the available plans page to subscribe to a plan before generating API keys. If you click see plans, on the left side, you can see search. That's what we want for $5 per 1,000 requests. Now, here's the thing though, it includes free $5 in credits every month. So, even though it asks you to start a subscription and provide a credit card, you can cap the usage at $5, which makes it so that you won't spend any more than the free $5 in credits they give you. To move forward, click get started. Click the checkbox to acknowledge the terms of use, then click subscribe, and enter your card information on the next page. After linking your credit card, you'll get a subscription successful confirmation and then you can click set usage limit and set the usage limit to $5 by clicking the apply button below here or entering it manually, then clicking create usage limit. So again, now even though our credit card is attached to this account, we won't spend any more than the free credits they give us. Click back to API keys on the left side, then click add API key. Give the key a name. We'll call it Paperclip. Link it to the search subscription and click add. That's the general process for how you get an API key. You make sure your account has credits, then you grab a key. We'll do the same thing for resend.com so we can send emails using our agents. Open a new tab, go to resend.com, click get started, create an account with Google, GitHub, or email. I'll use my email. Check your email to click the verification link. Here's what that looks like. Click confirm account, then click the confirm account button. And now we're into resend.com. Click add API key. And scroll down and click send email. If you go to your inbox, it might show up in spam, but there should be an email from resend saying congrats on sending your first email. Now that we've generated API keys for both Brave Search and resend, let's add those API keys to Paperclip so our agents can use those tools. Let's go back to Paperclip. And then let's give these tools to our CTO. On the left side, click on CTO. Click on the configuration tab, scroll down to permissions and configuration, and then find the environment variable section. Here, let's add the Brave Search API first. First, we'll create a key, which is essentially a name for our API key. It's a common format to use uppercase characters and underscores instead of spaces. So, we'll say brave_search_api. Go back to your Brave tab. And copy the API key. Return to Paperclip and paste in the API key in the value field. I've got this blurred out here because you don't want to share API keys with anybody. They're essentially passwords. If someone gets your API key, they can use it to access your account. Once you've pasted in the key, go ahead and click seal. You'll get a prompt asking for you to name this secret. By default, it gives it the same name as the environment variable, but with lowercase letters. So, you can go ahead and click okay. Now, instead of exposing your token here, you have a reusable secret you can grab. I'll show you how to reuse it in a minute. Next, let's add our resend key. In the second key field, type resend_api. Go to resend. Copy your API key. Go to Paperclip. Paste it in the value field and seal it. And then confirm the name. Don't forget to save the configuration. Now our CTO has access to both of these tools. Let's also give our founding engineer these tools. On the left side, click on the founding engineer, if you have one, or whatever other agent makes sense to give tool access to. Scroll down to configuration. And here for any second agent that you add, you'll still need to give it a name or a key. Again, we'll call it Brave Search API. But now, if you toggle from plain to secret, you can select the Brave Search API secret from the drop-down. And we can do the same thing for resend. And then click save to lock in the configuration. This is designed to work just like giving specific employees at an organization access to only the tools they need. If an employee doesn't need a tool, you don't give them that tool. But in this way, you can specify exactly which agent has access to exactly which tools. Which agent has access to which tools depends entirely on your company's goals and configuration. Now that our company's been organized and set up and certain employees or agents have access to tools, let's make sure that they can actually do some research and deliver a digest email. On the upper left, click new issue. And the new issue dialog window pops up. Now, Paperclip is not a chatbot. Like I showed you earlier, this interface functions more like a project management tool. To have your organization do things, you need to create a task or in Paperclip, they're called issues. The first manual issue we should create here is to make sure that the agents who have tool access can actually use them. So, we can title the issue something like test tools. Let's assign it by clicking assignee and select the CTO. If you want to, you can add a project. We can add it to the onboarding project here, which was created for us by default. And then we need to describe the details of the project. So, I'll say, I've given the CTO and founding engineer access to Brave Search API and resend API. You need to ensure that these agents can use these tools successfully. Both agents should run a test by doing some research, compiling a sample digest, and then sending me an email with the results. Note that any future agents that need access to these tools should be given access to them automatically. In the bottom portion of this issue creation dialog, you can set the status of this issue. To do is perfect for this. You can set the priority level. I'll set this to critical. And you can add additional context with labels and even file uploading if you need to give it photos or documents or something like that. Let's go ahead and click create issue. If we click over to our inbox, we can see that the test tools task is now live. And if we click on it and scroll down, we can see that the CTO is now reading through the task details and working on it. It even just updated the status to in progress. I'll give this just a few minutes to process, and I'll fast forward to the results when we have them. Now, while that's running, I want to point out that for more general tasks, I recommend actually assigning tasks to your CEO. For this test task, we knew exactly which agents should be responsible for testing these tools. But in the case that you want to do something big and organization-wide to achieve your goals, the CEO is the best agent because it can delegate smaller parts of that task to specific agents. But of course, feel free to assign tasks to whatever agent makes the most sense if you want to get a bit more granular and save some of the token usage that comes along with all of the delegation and thinking that the CEO and top-level agents do. Okay, some time has passed. I've fast-forwarded. And here in the inbox tab, you can see that the test tools issue has been moved to done. If we scroll down through the task, we can see the comments that the CTO added. And if we go to our inbox, we can see that we have an email from the CTO here and another email here, most likely from the founding engineer. So, both agents were successfully able to search the web and send emails. So, that's Paperclip. If you want the same setup, use the link on screen or the first link in the description to get started with a one-click Paperclip deployment template on a hosting or server with an extra 10% discount already applied and a 30-day money-back guarantee. Thanks for watching.