Everybody can build

You don't need to be a programmer to make websites, apps or gadgets anymore, and you don't even need an idea yet. All it takes is the wish to build something, and Claude does the heavy lifting. This page walks you through four steps, and our Telegram group helps you through every one of them.

What you need

1

Join the Telegram group

Do this first, before you install anything. Everybody gets stuck at some point, whether it's a setup step that doesn't go as planned or a build that won't do what you want, and the group is where you ask. Post a screenshot, what you pasted and what Claude answered, and someone will help you out.

Join t.me/everybodycanbuild
2

Set up Claude on your computer

Installing Claude Code is the only part you do outside Claude. After that, everything happens inside Claude: a few short commands to pick its settings, and then prompts that you copy and paste while Claude sets everything else up itself. Whenever it needs your hands for a login or a switch in Settings, it stops and tells you exactly what to click.

2.1
The only part you do by hand

Install Claude Code

Claude Code is Claude living in a terminal on your computer, where it can create files, run programs and set things up for you. Open a terminal: on a Mac press ⌘ Space and type Terminal, on Windows press Start and type PowerShell (not Command Prompt), and on Omarchy press Super Return. Then paste this line and press Enter.

curl -fsSL https://claude.ai/install.sh | bash

When it finishes, close the terminal window and open a fresh one so it notices the new program, then start Claude:

claude

Claude opens your browser so you can log in with your Claude account, and when it asks whether you trust the folder, say yes. If your Mac ever pops up a window offering to install developer tools, click Install and let it finish.

How to talk to Claude: type or paste what you want and press Enter. You can stop Claude at any moment with Esc. Next time you want to build, open a terminal and type claude, or claude --continue to pick up your last conversation where you left it.

Claude can't type your computer's password for you, so when something in the next parts needs it, Claude gives you a line to paste into a second terminal window instead, which is normal.

2.2
Type into Claude

Use Opus 5.5 on high effort

Opus 5.5 is the model this whole guide counts on, and high effort makes it think harder before it acts, which pays off when it's setting up your computer or building something real. Type these two commands into Claude one at a time, pressing Enter after each, and Claude remembers them for every future session.

/model opus
/effort high
2.3
Type into Claude

Give Claude Fable as an advisor

The advisor is a second, even stronger model that Claude consults at the important moments: before it commits to a plan, when it's stuck on the same error, and before it tells you it's done. Fable is the strongest Claude model, so it makes the best advisor, while Opus 5.5 keeps doing the everyday work.

/advisor fable
If Claude says Fable needs usage credits: on some plans, Fable is billed separately from your plan's included usage. Type /model fable once and accept, then /model opus to switch the main model back, and run /advisor fable again.
2.4
Paste into Claude

Turn on auto mode

In auto mode, Claude gets on with the work without asking you to approve every single step, while a safety check still reviews each action and stops anything risky. Press Shift Tab until the bottom of the screen says auto mode, and paste this so every future session starts that way.

Make auto mode my default permission mode for every session by setting permissions.defaultMode to "auto" in the settings.json file in my home folder's .claude directory. Keep everything else in that file as it is, and tell me when it's done.
2.5
Paste into Claude

Let Claude use your browser

With browser use, Claude can open websites in your Chromium, click around, fill in forms and check that the things it builds actually work. It uses the browser you're already logged into, so it can help with your real accounts.

Set yourself up to control my Chromium browser with the official Claude in Chrome extension, which works in Chromium too. Read https://code.claude.com/docs/en/chrome first so you follow the current steps. If Claude Code doesn't find Chromium on its own, set up the connection to it yourself.

Walk me through the parts only I can do, one at a time, and wait for me after each: installing the extension from the Chrome Web Store, signing in to it with the same Claude account I use here, and anything I need to type or restart in this terminal. If you need to be restarted, give me the exact line to type and remind me that "claude --continue" brings this conversation back. Turn browser use on by default so I don't have to do this again.

When it works, prove it: open a new tab, go to wikipedia.org, and tell me today's featured article.
2.6
Paste into Claude

Let Claude use your whole computer

Computer use lets Claude see your screen and use any app with the mouse and keyboard, like a spreadsheet, a photo editor or your system settings. On a Mac it's built into Claude Code, and on Windows and Linux Claude sets up an open-source tool that does the same job.

Give yourself the ability to see my screen and use my mouse and keyboard on this computer. Read https://code.claude.com/docs/en/computer-use first.

If the built-in computer use supports my operating system and plan, set that up. If it doesn't, find the best well-known, actively maintained open-source computer-use tool for my operating system and connect it to yourself instead, and tell me what you picked and why before you install it.

Tell me exactly what I have to do myself, one step at a time, and wait for me after each, such as menus to open in this terminal or permission switches in my system settings. If I have to quit and reopen the terminal, tell me first that "claude --continue" brings this conversation back.

When it works, prove it: open the calculator app, work out 1234 × 5678 by clicking its buttons, and read me the answer from the screen.
While Claude drives: it sees whatever is on your screen, so close your banking and password windows first. You approve each app it wants to use, and Esc takes control back right away.
2.7
Paste into Claude

Give Claude passwords safely

Sooner or later Claude will need a password or an API key, for example to publish your website or log in to a service. Pasting those into the chat is a bad habit, because the chat is saved and you lose track of where your secrets went. Instead, Claude sets up Vaultwarden, a password vault that runs on your own computer with a web page where you add passwords, and then fetches secrets from it straight into the programs that need them.

Set up a password vault for you on this computer with Vaultwarden, running locally with its web interface so I can add passwords in my browser. Follow these rules the whole time: never ask me to type a password into this chat, and never print a secret's value, in your messages or in command output. Whenever something needs my computer's password, give me the exact line to paste into a second terminal window.

1. Install Docker if I don't have it yet, and walk me through any part I have to click through myself.
2. Run Vaultwarden in Docker so it only listens on this computer, keeps its data in a folder in my home directory, and starts again by itself after a restart.
3. Open its web vault in my browser and have me create my account there. Remind me to pick a master password I don't use anywhere else and write it on paper, because nobody can recover it. Once my account exists, turn off new sign-ups so nobody else can register. Then have me add an item named "test" with a dummy password like hello123.
4. Back up the Vaultwarden data folder automatically every day to a second place I choose, such as a cloud-synced folder, because if that folder is lost, every password in it is gone.
5. Install the official Bitwarden CLI ("bw") and point it at my local Vaultwarden.
6. Write a small unlock script that I run myself in a separate terminal window, and make sure it runs on my operating system without extra fiddling. It logs in and unlocks with hidden prompts, then saves only the session key to a file in my home folder that only my user can read.
7. Write a helper command that takes an item name, an environment variable name and a command, and runs that command with the item's password in that variable, without ever printing it or putting it on a command line. If the vault is locked, it should say so and tell me to run the unlock script.
8. Add short instructions to the CLAUDE.md file in my home folder's .claude directory so every future Claude session knows to use this helper for secrets, never prints them, and asks me to add any secret it needs to the vault.
9. Test it with the "test" item and show me only the number of characters in the password, then give me the vault's address so I can bookmark it.
Using it later: when Claude says it needs, say, a Cloudflare token, open your bookmarked vault page, add the token under the name Claude suggests, and tell Claude it's there. If the vault has locked itself, run the unlock script again. Only put secrets in this vault that you're happy for Claude to use, because keeping them out of the chat doesn't stop Claude from using them.
2.8
Paste into Claude

Give Claude a home in tmux

tmux keeps your Claude sessions running in the background, so closing the terminal window or losing the connection on your phone doesn't stop anything. You get tabs for several Claude sessions at once, and a tab turns yellow when the Claude inside it is waiting for you.

Install tmux and set it up for working with you all day. If this computer can't run tmux natively, use the closest setup that works here. Whenever something needs my computer's password, give me the exact line to paste into a second terminal window.

1. Typing plain "tmux" attaches to one persistent session called "main", creating it if it doesn't exist yet.
2. The mouse works: the wheel scrolls back through a long history (50,000 lines), and dragging or double-clicking copies to my own clipboard, even when I'm connected over SSH from another device. Double-clicking a link copies the whole link without trailing punctuation.
3. Keys like Shift+Enter reach you inside tmux, so multi-line prompts still work.
4. Tabs are numbered from 1 with no gaps. Ctrl+Shift+T opens a new tab in the current folder, Ctrl+Shift+W closes one after asking, Ctrl+Shift+Left and Right switch tabs, and Ctrl+Shift+D detaches while everything keeps running.
5. Each tab is named after what the Claude session in it is working on, and the active tab is highlighted.
6. A tab turns yellow when the Claude session in it has finished or needs my input and I'm not looking at that tab, and goes back to normal when I switch to it or reply. Use Claude Code hooks in my user settings for this.

Write a short explanation of each setting as a comment in the config. Then open a tmux session yourself to check that it all loads without errors, and show me a cheat sheet of the shortcuts.
2.9
Paste into Claude

Build from your phone, anywhere

Tailscale puts your computer and your phone on a private network of your own, so you can connect to the computer from anywhere without opening it up to the internet. SSH is how you connect, and once you're in you just type claude like you would at home. This works from another laptop too.

Set me up so I can connect to this computer from my phone or another computer, from anywhere, and keep working with you there. Use Tailscale and SSH, and walk me through the parts only I can do, one at a time, waiting for me after each.

Whenever something needs my computer's password, give me the exact line to paste into a second terminal window, or tell me what to click.

1. Install Tailscale on this computer from tailscale.com/download (on a Mac use the Standalone app, not the App Store one) and have me log in when you're ready.
2. Turn on an SSH server for my user only. On a Mac that means I switch on Remote Login in System Settings, on Windows the built-in OpenSSH Server, and on Linux Tailscale SSH. Make it accept SSH keys only, never passwords, and if you can, only connections that come through Tailscale.
3. Make every SSH login drop me straight into my tmux "main" session, so my work is waiting for me and keeps running if my phone loses signal. If tmux isn't set up yet, set it up first.
4. Stop the computer from going to sleep while it's plugged in, so I can always reach it. If it's a laptop, tell me the lid has to stay open, and tell me whether disk encryption means someone has to log in at the computer after a restart.
5. Tell me which apps to install on my phone: the Tailscale app, logged in to the same account, and a free SSH app such as Termius. Help me make an SSH key in that app and send you only its public key, then add that key to this computer. Give me the exact host name and username to type in, and have me save it as a host called "work" so connecting is a single tap.
6. If I use other computers, give me one command to paste on each of them that installs Tailscale if needed, makes an SSH key, and adds a "work" shortcut, so typing "ssh work" there connects me straight into my tmux session here. Add each computer's public key to this one when I send it to you.
7. SSH into this computer yourself the same way my phone will, and check that I land in tmux, that claude starts there and that it's still logged in. Fix whatever doesn't work, such as programs missing from the path in SSH sessions.
8. If I have a Vaultwarden vault on this computer, make its web page reachable from my phone through Tailscale only, over HTTPS, so I can add passwords from anywhere.
9. Save a short cheat sheet to my home folder as remote-cheatsheet.txt with how to connect from my phone and from other computers, and what to type once I'm in.

When everything is set up, wait while I connect from my phone and tell you it worked.
Only want to chat from your phone? There's a simpler way for that. Inside Claude, type /remote-control, then open the Claude app on your phone and tap Code to keep talking to that same session. The SSH setup is for when you want the full computer from anywhere.
2.10
Paste into Claude

Teach Claude how we build

Claude remembers rules you give it across every future session. These are the habits our group builds with: fast websites that look great when shared, cheap servers you own instead of pricey hosted platforms, and backups for anything you can't get back. Claude will ask you to create the accounts it needs, like GitHub, Cloudflare or Hetzner, the first time it needs them, and their keys go into your vault.

Remember these rules for every future session by adding them to the CLAUDE.md file in my home folder's .claude directory, written as clear instructions to yourself:

- When you build a website, always make a great 1200×630 share image with full Open Graph and Twitter tags, and push for a Lighthouse score of 100, especially for SEO. Never write staccato sentences (runs of clipped fragments like "Fast. Simple. Yours.") in the copy. Avoid server-side rendering: build a single-page app and prerender it for SEO whenever possible. Keep the bundle as small as you possibly can, and put the site behind Cloudflare's proxy and CDN so it loads fast everywhere.
- If a website can be static, keep it static and host it on GitHub Pages.
- Don't use Vercel, Netlify or other hosted platforms. When something needs a server, find a good cheap one on the Hetzner server auction (hetzner.com/sb) and set it up yourself as much as possible.
- When you deploy to a Hetzner server, manage the services on it with Tugport (tugport.com), and always push me to set up sensible backups of important data that can't be recreated.
- Use Cloudflare for domains: buying them, managing them, DNS and everything else.
- Use Mailgun on its free tier for sending email.

When you've saved them, read the rules back to me as a short list.
3
The hardest step

Collect ten ideas

This is the toughest step, and the one most people skip. Ideas rarely show up when you sit down and try to think of one, but they hide in the little annoyances of your day: a chore you repeat every week, a paper list on the fridge, an app that almost does what you want, a question you keep googling.

So start looking out for those pain points. Whenever you notice one, pull out the notes app on your phone and write the idea down right away, no matter how ambitious or silly it sounds, because the silly ones are often the best to build. Keep going until your list has ten ideas, and only then pick the one you're most excited about.

The best ideas are as close to a single-button app as possible: you open it, press one button, and something useful happens. An idea that small is one you can actually finish, and one you'll keep using every day.

My ten ideas
  1. A button that tells the family group whose turn it is to cook
  2. A button that turns a photo of a receipt into a row in my tax spreadsheet
  3. A gadget that texts me when the mailbox opens
  4. A button that gives my kid the next multiplication question
  5. A button that books my usual haircut slot for next month
4

Build it

Pick one idea from your list, and build it together with the Telegram group rather than alone. The easiest way to waste a month is to build in silence and only find out at the end that the idea needed a tweak, so we keep each other honest with three habits.

Build it with the group

  1. Pitch it before you build it. Post your idea in the group in a couple of sentences: what the button does and who presses it. Someone will help you make it simpler, spot a problem early, or point you to a shortcut.
  2. Check in every time you have something to show. A screenshot, a link or a ten-second video is plenty, even when it's rough. Showing your progress often is what keeps you from spending weeks building into the void.
  3. Ask the moment you're stuck. We're in there all the time, and a question that eats your evening is usually a two-minute answer for someone who has hit it before.
Pitch your idea in the group

Then tell Claude about your idea the way you'd tell a friend: what it should do, who it's for and what "done" looks like. You can build literally anything this way.

You don't need skills or prompt engineering. You'll hear about skills (SKILL.md files you install into Claude), plugins and long, elaborate prompt templates that promise better results. Skip them all for now and build with stock Claude Code on Opus 5.5 at high effort, describing what you want in plain words. Once you've built a thing or two, feel free to experiment, and you'll most likely find that the extra skills and fancy prompts make the result worse rather than better, and drop them anyway.

Tell Claude what it must ask you about before doing it, mention your browser or computer use when you want it to click around for real, and ask it to check its own work, because it is good at that. Here are a few prompts to get you going:

WebsiteMake me a one-page website for my bakery with our hours and menu, then open it in my browser and fix anything that looks off on a phone-sized window.
Phone appBuild an Android app that keeps track of whose turn it is to do the dishes, and walk me through installing it on my phone.
HardwareI have an ESP32 board and a soil moisture sensor. Tell me how to wire them, write the code, and make it message me on Telegram when my plant needs water.
BrowserUse my browser to find three flights to Lisbon for the second week of March under $500. Put them in a table, and don't book anything.

And when your first version works, even if it's tiny, post it in the group. That's the moment everybody there is waiting for.