First, congratulations. Let me show you exactly what you have, whether you purchased the SBOS Website Operator or you picked up the free, open-source version of the website.
Everything about this website is centred on the AI receptionist that powers it: the agent that talks to the people who land on your site, using the information from your site. Behind that sits a set of tools that feed both the website and the receptionist at the same time. Update once, and both change.
Those tools cover your people (the team, the about page, the booking page per person), your products and services (names, prices, bundles, offers), your company details (the name on the incorporation papers, address, opening hours, phone, email, WhatsApp, Instagram) and your company knowledge (every page and every blog post on the site). In our world we call this the backend.
Here is what that means in practice. The people on the team page are loading from the backend, not from the code. Change who is on the team and the website updates, and the receptionist knows who has joined and who has left, without you opening a single file. The company details in the footer come from the same place. Change your phone number and it changes on the site and in the receptionist together. Products work the same way: update a price, a name, a bundle, and it lands on the pricing section and in every conversation the receptionist has from then on.
That is the big benefit of the SBOS Website Operator: one update reaches your receptionist and your website at once. On top of that, the operator is a companion that designs, rebrands, rewrites and changes anything you can imagine about this site. No developer, no fixed monthly bill. All you need is a Claude subscription (or Gemini, or whichever you use); I strongly recommend Claude Code, and it is what I use for every step below.
To show you what the end result looks like, I ran this whole process for a real client, Thai Qi Holistics. The operator rewrote the copy, the sections and the flow diagram, described the conditions they help people with, laid out their treatment process and consultation approach, and put their actual products and prices on the pricing section. Every one of those changes was made by the agent, in conversation.
If you took the free version
You get the design and the components: all the code, free, yours to own. What you do not get is the functionality, because that comes from Second Brain OS. The chat (both the full-width one on the home page and the widget on the side) will not load. The people, the products and the company details stay as dummy data. Forms, analytics and the knowledge base are not there either.
You can absolutely clone the repo, use the design, build your website and own the entire code. Keep watching regardless, because you will see every part of the site come to life, and you can upgrade at any point. The same receptionist that answers on your website powers your WhatsApp and Instagram in a single click.
The operator is the part that makes the design do something: it acts as a done-for-you developer you buy once and work with as long as you like. Spend a good hour today (maybe even half of it) and you will have a fully functioning, well designed website that loads instantly.
Step 1: GitHub and Vercel accounts
Two accounts, both free.
GitHub is where the code of your website lives. Head to github.com, create an account or sign in (I signed in with Google for the demo account).
Vercel is where the website runs. Go to vercel.com, press Sign up, choose Continue with GitHub, and pick the GitHub account you just signed into. Choose personal projects, pick the Hobby plan, and name the workspace whatever you like: "My company", "My business", "My Second Brain OS". Do not create any projects yet; leave it empty.
Step 2: Press Deploy
Either on this page or on the open-source repository you will find a deploy button: the Vercel logo (a white triangle in a black circle) next to the word Deploy.
Press it. It picks up your Vercel account, then:
- Select GitHub as your git provider and press Add Git scope.
- If this is a brand-new GitHub account, choose All repositories. If you are tech-savvy, clone the repo into your own GitHub account first and select only that repository.
- Name the project. "My startup", "My website", your company name, anything.
- Press Create.
A few minutes later your website is live. If you want to watch it happen, open the build logs. Then right-click the domain Vercel gives you and open it in a new tab.
You now own a fully functioning, live website. Whether or not you purchase the operator, this part is yours.
What you will also notice: the chat does not load, entering a name and email on the home page does not open the receptionist, and the company details, people and products are all dummy data. That is exactly why the next step exists.
Step 3: Open Claude Code and call the Website Operator
I am assuming you installed Claude Code and connected Second Brain OS to it in the earlier guide, and that you have most likely already set up your basic AI receptionist through it. If not, go back and do that first; it is very simple.
Create a folder for the website. On a Mac, open Finder, press Cmd+Shift+G, type ~ (the key under Escape, with Shift) and press Enter to land in your home folder. Create a new folder and call it my-startup (or the same name you gave the Vercel project).
Open Terminal (Cmd+Space, type Terminal, press Enter) and go into the folder:
cd my-startup
claude
Say yes when it asks if you trust the folder. Type /mcp and confirm that Second Brain OS is connected. Then type website operator, choose the SBOS Website Operator agent, and send it a prompt like this, with the domain Vercel gave you pasted in:
Hello. I have just made my GitHub and Vercel accounts, after which I took my website live to https://my-startup.vercel.app. I need your help setting it up.
Two settings worth changing while it works. Type /model and choose Opus; Sonnet will do fine, Fable is overkill. And press Shift+Tab to change how much it asks you: Auto mode lets it continue without asking permission, Accept edits lets it edit files but asks before anything else. I used accept edits for the recording. You can press Escape at any point to interrupt it.
Step 4: Connect GitHub on the command line
The operator will first check whether your computer is signed into GitHub, and with which account. Nearly everyone at this point is not, or is signed in as someone else. It asks one question at a time:
- Your GitHub username. You will find it at the top of GitHub, or in the URL of the repository Vercel just created for you (refresh GitHub and my-startup will be there; that URL includes your username).
- Whether to log out and switch if a different account is signed in. Say yes.
- Where to pull the code. Choose the my-startup folder you are already in.
It then gives you a login command to run yourself. Copy it (without the leading exclamation mark), press Cmd+T to open a new terminal tab, paste it and press Enter. A short code appears. Copy the code, press Enter, pick your GitHub account in the browser, paste the code, press Continue and Authorize GitHub. That is it: you are all set.
What just happened: you set up the git command line, which is what allows the operator to work directly with the code you copied from me, which is now the code of the website you own.
Go back to the first tab and tell it: thanks, all done, what's next? It verifies it is logged in as the right account, clones your website, and pushes a small, harmless change to prove the whole pipeline works. You can verify that yourself in Vercel under Deployments. If the deployment shows as blocked, do not worry; the operator resolves that on its own.
Step 5: Give the operator eyes on Vercel
Next it installs what is called the Vercel MCP: a connection that lets the operator see whether the changes it makes actually go live, read build logs, and fix a failed deployment itself.
It runs the install command for you, then asks you to do one thing it cannot: restart Claude Code so the new connection loads.
- Wait for it to finish talking (its replies can be long).
- Press Ctrl+C twice to exit.
- Type
claudeto start again, then/resumeand pick this conversation. - Type
/mcp. Vercel is listed and needs authentication. Arrow down to it, choose Authenticate, and a browser tab opens. - Select your workspace and your project (my-startup), press Allow. Authentication successful.
Back in the terminal, tell it done. It checks that its earlier change went through, makes one more small change to confirm everything is smooth, and you can open the domain in a new tab to see the site still loads.
Three things are now checked off: the operator is running, it is signed into GitHub as you, and it can see your live deployments.
Step 6: Your own domain
You want the website on your own domain name, and you want the chat live (both the full-width receptionist and the side widget). You are in one of two situations.
You do not have a domain. The easy path: open your project in Vercel, go to Domains, and buy one there. It links to the project automatically.
You already have a domain. Log into wherever you bought it (GoDaddy, Namecheap, anywhere), find the domain and copy it. In Vercel, click out of the project into your workspace, go to Domains, press Connect External, paste the domain (watch for trailing spaces; I made that exact mistake on camera), press Continue, select your project and Add Domain.
Vercel now asks you to update the domain's configuration. It offers an A record, but what I recommend is Vercel DNS: copy the two nameservers Vercel shows you, go to your registrar, open the domain's Nameservers setting, choose custom DNS, paste both and save. From then on everything about your website is managed in one place instead of split between the registrar and Vercel.
This usually takes three or four minutes to update, but occasionally it takes well over 24 hours (registrars say up to 48). Once it is through, Vercel generates an SSL certificate, which is what makes your site https rather than http: whoever visits, their connection to your website cannot be intercepted. Then the domain connects to your project. Check it under the project's Domains, and open the domain in an incognito window to see the site load.
Step 7: Register the domain with Second Brain OS
Now the same domain goes into the Second Brain OS AI agent dashboard and the Second Brain OS CRM. If you came through the funnel you have probably already registered your domain and an OpenAI public domain key; if so, you can skip the dashboard part, or do it again to be safe.
The OpenAI domain key. Go to OpenAI's domain allowlist, log in, and add your domain: the root domain only, no https://, nothing after it, spelling triple-checked. OpenAI gives you a public domain key for it. That key is what lets the chat run on your site from the browser without exposing anything private.
Proving the domain is yours. Second Brain OS will only save a domain once you have shown you control it, so no one can register your domain against a different account. Add one record wherever your domain's DNS is managed (Vercel DNS if you moved the nameservers, otherwise your registrar): type TXT, name _sbos-verify, value your Second Brain OS user ID (the operator will tell you what it is). It usually propagates in minutes.
Saving domain and key together. In the Second Brain OS dashboard, paste the domain first, then the domain key, and press save. They have to go in together.
The CRM inbox. At the bottom of the dashboard press Deploy to CRM. The Second Brain OS CRM opens. Go to Settings → Inboxes → Add inbox → Website, name it after your domain, press Create, assign yourself to it, and you get a script snippet.
Now give the operator all three in one message, and it does the rest:
My domain is thaiqiholistics.com, the domain key is domain_pk_…, and from the Second Brain OS CRM here is the script: [paste the whole snippet]
The operator only needs the token inside that snippet; there is no code to change. Behind the scenes it also writes the correct Second Brain OS endpoints into a local .env.local file for your site.
Step 8: Answer the interview
From here the operator asks you one question at a time and gathers what it needs to replace every piece of dummy data with a real source:
- Your business name and a little about the business.
- Core company details: physical address, contact phone, email, opening hours.
- Who you are, or whoever should be the face of the business: a short bio and a photo. If you are the only person in the company, that is fine; your about page comes from your employee profile.
- Your services, and any digital products (e-books, courses, bundles), with real prices and a currency.
Wherever it is confused it asks. When it asked me whether eight named courses should each become a separate catalogue entry, the answer was yes, separate entries. Pick the currency (USD in my case) and submit.
Step 9: Your one manual task, the API key
Once the setup is done there is one tiny thing only you should do. Log into secondbrainos.com, open your dashboard and copy your API key.
Open your my-startup folder in Finder and press Cmd+Shift+. to reveal hidden files. You will see .env.local. Open it with TextEdit (Notepad on Windows). Everything is filled in except one line: SECONDBRAINOS_API_KEY. Paste your key there, save, close.
The operator leaves this line empty on purpose. A lot of people paste their API key into the chat with the agent, and that is strongly not recommended. The file is the safe place.
Tell the operator: it's done, show me the website now. It starts the site on your own computer and opens it in your browser for you. Look around. The footer shows your real company details and phone number. The people page shows you. The products page shows your actual products. The about page and booking page are yours. The blog loads. Click the chat, say hello, and the receptionist answers.
One note on that last point: the chat only works if you set up your basic AI receptionist in the earlier step. If you did, it works out of the box; you only had to provide the API key.
Once you have looked yourself, let the operator investigate to its own satisfaction. It checks the same things with its browser tools and flags anything that is not working.
Step 10: Upload the environment file to Vercel
The site works on your computer. Now the live one needs the same connections.
In Vercel, go Home → your project → Settings → Environment Variables → Add. Drag the .env.local file straight into the box; it picks up everything. Two entries you can delete safely: the GitHub repo and GitHub token variables. Keep the rest; that is what makes the receptionist answer and your team appear.
Press Save, then Redeploy (if you miss the prompt, it is under Deployments). A minute later the status changes to Ready. Open your domain in a new tab and the real data is flowing on the live site: your address, your profile, your products.
Step 11: The single long pass
Your website is up. Now the copy has to become your business rather than mine, and this is the fun part. Tell the operator:
I have updated the environment variables on Vercel. Now I need you to update the copy of the homepage based on the info I gave you. Feel free to ask me any questions as you go along, don't hesitate.
It rewrites the site in one long pass, like a developer making sure everything lines up, without you being bogged down in the technical details. This takes a while: five or six minutes for the rewrite, sometimes more. Be patient with it, and with yourself.
Watch it land under Deployments ("Rewrite the home page"), and when the status flips from Building to Ready, open your domain. For Thai Qi Holistics the entire home page came back on-brand: the headline, the paragraph about a clinic and education centre grounded in medical qigong, a diagram of how treatment, teaching and rehabilitation work together, the conditions they treat, and how treatment is shaped around the person rather than a template.
Was it perfect? No. An image here, a card there that did not fit their flow. But that is now your job as the human: look at the site, and give the operator feedback based on what you see. Start to finish took me fifteen to twenty minutes of setup and maybe fifteen more of feedback to get it where I wanted it.
Analytics, out of the box
With the operator purchase, analytics come enabled by default. Visit secondbrainos.com/dashboard/analytics: daily visits, where people come from, how long they stay, which pages they read, which locations and browsers, and every event as a row in a table. Leads and purchases show up there too.
One more thing, and it matters. We live in a world more privacy-conscious than ever, and I think user privacy should be everyone's priority. For that reason the analytics are anonymous. If you want to tie activity to a real person (what form they filled, which products they showed interest in, whether they booked) you need two things: their consent, and them declaring who they are, for example through a form. With those, you see their activity against their contact record in the CRM. Without them, you still get the full anonymous picture.
So you need a cookie consent banner from day one, not after a legal notice. I have done the wiring; your only job is a CookieYes ID. Create an account at cookieyes.com (the basic plan is about $10 a month with a free trial), connect your domain there the same way you connected it to Vercel and Second Brain OS, and copy the ID it generates. Paste it into the Analytics tab of the Second Brain OS dashboard and save. Open your site in an incognito window and the banner appears.
We go much deeper into consent and identity in the forms and data guide. For now, add the ID and let people choose.
You now have a fast, well designed website on your own domain, with an AI receptionist that answers from your real data, analytics you can act on, and an operator agent that will keep designing, rewriting and shipping changes for as long as you own the site.
