Big Idea

The argument in one line.

Paperclip reframes a team of AI agents as a literal company org chart, where a CEO agent autonomously hires and delegates to C-suite agents against a stated mission, shrinking the human's role to approving hires, setting budgets, and reviewing blocked work.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You're curious how multi-agent AI systems get structured beyond one chatbot answering one prompt at a time.
You want to understand the heartbeat/poll-loop pattern that lets agents act autonomously over time instead of only on-demand.
You're evaluating whether to self-host an open-source agent orchestration tool on your own server.
You want a concrete example of scoping tool credentials (search, email) per-agent instead of sharing one global key.

SKIP IF…

You want a no-code, consumer-friendly AI assistant, not something that requires a VPS, Docker, and a terminal.
You're looking for proof the agents produced business value, not just a test email reaching an inbox.

TL;DR

The full version, fast.

Paperclip models an AI workforce as a company: you give it a mission, hire a CEO agent, and the CEO hires its own C-suite (CTO, CMO) and sub-agents to break the mission into tasks. Each agent runs on a heartbeat, a timed wake/check/sleep loop, instead of only responding to prompts, so the company keeps working around the clock on a VPS. New hires default to needing human approval, tool access (search, email) is granted per agent as scoped secrets, and tasks move through a Kanban-style pipeline so a blocked task is the one signal a human actually needs to act on. The demo closes with two agents independently completing a task and emailing the result, the concrete proof the loop works end to end.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:29

01 · Intro

Cold open framing the end state: a working AI company that sends real output to your inbox.

00:29 – 02:25

02 · What Is Paperclip?

Defines Paperclip as a company-of-agents tool: a CEO agent hires C-suite agents, each agent works on a heartbeat loop, and the whole thing runs on a server around the clock.

02:25 – 08:06

03 · Get Your Server Running

Deploys Paperclip via a one-click Hostinger VPS template, then connects an OpenAI Codex account to the server through terminal device-code authorization.

08:06 – 17:32

04 · Create Your Company

Names the company, sets its mission, configures the CEO agent's adapter and model, gives it a first task, then tours the live dashboard: inbox, org chart, issues, and the Kanban board as the CEO starts hiring a CTO and CMO.

17:32 – 22:41

05 · Give Your Agents Tools

Connects Brave Search and Resend as external tools, stores the API keys as sealed secrets, and grants scoped access to the CTO and founding engineer agents.

22:41 – 25:48

06 · Put Your Company to Work

Assigns a test-tools issue to the CTO, the agents research and send real emails, and the video closes on the sponsor's discount and guarantee.

Atomic Insights

Lines worth screenshotting.

Paperclip models an AI workforce as a company: a CEO agent hires C-suite agents who delegate further, instead of one chatbot handling one task at a time.
Every agent runs on a heartbeat: it wakes on a timed interval, checks the mission and its tasks, works if there's something to do, then goes back to sleep.
New agent hires default to requiring human approval, making the human operator the 'board of directors' who signs off before anyone joins the org chart.
An agent company has to run on a server, not a laptop, because the agents only keep working while the machine stays powered on.
Authenticating an automation system with a subscription login instead of a metered API key can violate that provider's terms of service, a real distinction when chasing the cheaper option.
Tool access like search and email is granted per agent as scoped environment-variable secrets, the same least-privilege pattern as giving one employee only the tools their job needs.
Tasks in Paperclip are called issues and move through a Kanban pipeline: to-do, in progress, in review, blocked, done, canceled, the same lifecycle a human project tool uses.
A blocked issue means an agent hit an external limit, like missing permissions or access, which makes that column the one a human actually needs to watch.
Routing a narrow task directly to a specialist agent skips the extra tokens a CEO agent spends delegating and reasoning before work even starts.
The entire chain, from VPS to connected tools to a completed task, produced one verifiable external result: two agents independently researched and sent a real email to the operator's inbox.

Takeaway

How an AI agent company stays accountable

AGENT SYSTEMS

Paperclip structures AI agents as a company org chart, and the transferable lessons are about how autonomous multi-agent systems get scoped, scheduled, and kept answerable to a human, not about the specific tool.

02What Is Paperclip?

Multi-agent orchestration tools are starting to model AI teams as literal company hierarchies instead of a flat set of bots, with a CEO agent delegating down an org chart.
The core value isn't a more capable model, it's a scheduling loop (heartbeat) that lets agents act autonomously over time instead of only responding to a single prompt.
Open-source tools built around a fast-moving AI stack ship breaking changes constantly, so 'still early software' is a real caveat, not just marketing modesty, when adopting one.

03Get Your Server Running

Running an always-on agent system requires a machine that stays powered independent of your own device, which is the actual reason a server becomes necessary, not a feature preference.
Device-code authorization, pairing a terminal-generated code with a browser login, is becoming a standard way automation borrows access to a subscription account without exposing raw credentials.
Using a subscription login instead of a metered API key to power automation can cross into a terms-of-service violation with the provider, a distinction easy to miss when chasing the cheaper option.

04Create Your Company

A single high-level mission statement, not detailed instructions, is what an autonomous agent system needs to start generating its own task breakdown.
Letting an AI hire further agents without human approval trades oversight for speed; the tool defaults to requiring approval because unchecked hiring is the riskiest autonomy setting.
Tracking AI work as issues moving through a to-do/in-progress/blocked/done pipeline borrows directly from human project management, which makes a multi-agent system auditable instead of a black box.

05Give Your Agents Tools

Planning and reasoning are table stakes; agents only become useful once they're handed real external tools, like search and email, with their own credentials.
Scoping each tool credential to a specific agent, rather than sharing one global key across the system, limits the blast radius if one agent misbehaves or leaks.
A hard spending cap on a third-party API is a practical safety rail for letting an autonomous agent spend money on your behalf.

06Put Your Company to Work

The real test of an agent system isn't whether it can plan, it's whether a task assigned at the top produces a verifiable external artifact.
Routing a specific, well-scoped task directly to a specialist agent is more efficient than routing everything through the top-level agent, which spends extra tokens delegating before work starts.
A blocked task is the one status that reliably needs a human's attention, since it means the agent hit a limit, like missing access, that it can't resolve itself.

Glossary

Terms worth knowing.

Heartbeat: The interval at which a Paperclip agent wakes up, checks its assigned tasks and the company mission, does any available work, then goes idle again.
Adapter (agent adapter type): The setting that connects one Paperclip agent to an AI provider, such as Claude Code or OpenAI Codex; a single agent can only use one adapter even though a company can mix providers across agents.
Issue: Paperclip's unit of work, equivalent to a task or ticket, assigned to an agent and tracked through a pipeline from to-do to done.
Org chart (agent hierarchy): The visual map of which agents report to which, with a CEO agent at the top delegating to C-suite agents and their own sub-agents.
Device code authorization: A login method that lets a service authenticate from a server's terminal by pairing a one-time code typed there with a sign-in completed in a separate browser tab.
VPS (virtual private server): A rented, always-on slice of a remote server, used here so an agent company keeps running continuously instead of stopping when a personal computer is closed.

Resources

Things they pointed at.

02:18toolHostinger VPS (affiliate link) ↗

00:00toolPaperclip (open-source AI org-chart agent tool)

04:17toolOpenAI Codex / ChatGPT ↗

18:11toolBrave Search API ↗

19:28toolResend (email API) ↗

Quotables

Lines you could clip.

00:30

“If an AI agent is an employee, Paperclip is the company.”

tight one-line thesis metaphor, needs no setup→ TikTok hook↗ Tweet quote

00:53

“The agent is one worker. Paperclip is the whole office. The org chart, the goals, the budget, the rules.”

extends the metaphor with concrete specifics→ IG reel cold open↗ Tweet quote

11:31

“All of the employees are agents. You, the human, are the board of directors.”

reframes the human's role in one line→ newsletter pull-quote↗ Tweet quote

25:33

“If you want the same setup, use the link on screen... an extra 10% discount already applied and a 30-day money-back guarantee.”

shows the exact shape of the closing sponsor pitch→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

metaphor

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 organized to themselves and produced 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 KVM2 plan to the KVM1 plan. KVM1 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 MedicsMedia 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 one month, 12 months, or 24 months. You need to choose at least 12 months to take advantage of the coupon. If you choose one 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 because 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 ChatGPT .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 Codecs 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 ChatGPT 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 forward slash bin forward slash 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 at server followed by a number to node at 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 dash dash device dash.

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 ChatGPT account.

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 ChatGPT account. 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 ChatGPT tab 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 in to 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 TechCo.

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 Cloud Code via API earlier, you can leave it set to Cloud 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 past 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 roadmap 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 cancel 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.

The 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 modal.

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 slash search slash 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 the 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'll 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 Variables 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 OK. 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 underscore 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 dropdown. 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 needed to give 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.

The Hook

The bait, then the rug-pull.

Paperclip pitches itself as a company you run with AI instead of people: give it a mission, hire a CEO agent, and watch that agent hire its own team and start delegating work without being told what to do next.

Frameworks

Named ideas worth stealing.

00:53model

CEO to C-suite to sub-agent org chart

AI agents are structured like a company hierarchy: a CEO agent reads the mission and hires specialized C-suite agents (CTO, CMO), which can in turn hire and delegate to their own sub-agents.

Steal formodeling any multi-agent system as a reporting hierarchy instead of a flat swarm of equally-privileged agents

15:41concept

Heartbeat wake-check-sleep loop

Each agent polls for work on a configurable timed interval, acts if there's something assigned, then goes idle, which is how the system stays running continuously without a human triggering every step.

Steal fordesigning any autonomous agent loop around a bounded poll interval instead of a tight always-on loop

20:10concept

Scoped per-agent tool secrets

External tool credentials are sealed once as reusable secrets, then granted individually to only the agents whose role needs them, instead of one shared key every agent can see.

Steal forcredential scoping pattern for any multi-agent system that shares a small set of API keys

CTA Breakdown

How they asked for the click.

VERBAL ASK

25:33link

“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.”

Classic affiliate-tutorial close: the same Hostinger discount and guarantee pitched once at the 2-minute mark is repeated verbatim as the sign-off, bookending the whole video with the sponsor link.

MENTIONED ON CAMERA

02:18toolHostinger VPS (affiliate link) ↗

04:17toolOpenAI Codex / ChatGPT ↗

18:11toolBrave Search API ↗

19:28toolResend (email API) ↗

Storyboard