I Built and Launched a Website with Bolt.new in 5 Minutes

I had a domain called salonmate.nl sitting around with nothing on it. I opened Bolt.new, fed its new Forge agent a copy file and a design document, and launched a working landing page with a waitlist database on that domain in five minutes.

The one insight worth copying is this: write your copy and your brand rules before opening an AI builder. If you ask an agent to invent the text, choose the palette, structure the layout, and write the code all at once, you get generic garbage. When you give it the text and the design tokens up front, it only has to do what it is actually good at: assembling the code.

Why Bolt Forge brought me back

Bolt.new launched in October 2024, before anyone was calling this vibe coding. I used it early on because it was a fast way to spin up prototypes and small sites without touching local setup. Over time they added integrated hosting, an integrated Supabase database, and slide generation using your branding.

I went back because StackBlitz announced Bolt Forge. Forge runs open models on Bolt’s reserved infrastructure and gives up to 50 times more usage allocation compared to the standard setup. There is a $9 sign-up offer that runs until October 14. What the pricing and the up to 50x Pro allocation look like after that date is not a given.

There is a clear trade-off. To use Forge, you must consent to let StackBlitz use your session data to train their models. That includes runs, generated code, project files, and error or fix traces. If you do not want your code used for training, you can use the other Bolt agents instead.

Bringing my own copy and DESIGN.md

The project was SalonMate, a booking tool for hairdressers, barbershops, and salons paired with a WhatsApp AI agent to handle appointments and reminders. I bought the domain months ago and kept putting off building the site.

Before opening Bolt, I prepared two files. I drafted the website copy in a markdown file using Gemini. Then I used Codex to write a DESIGN.md file containing the brand guidelines, font choices, and colors I wanted for SalonMate.

Help me build SalonMate first.
First, build all the pages.

I selected Kimi K3 inside the Forge model picker. I deliberately avoided giving technical constraints like asking for Astro or specifying a framework. I dropped in the copy file and DESIGN.md and let the agent make the architecture choices.

The agent built a working stack in one pass

My dashboard showed 10 million tokens available on my plan, plus some extra tokens left over from an earlier purchase. Once I gave consent for Forge data collection, the agent started building.

It set up the fonts and colors from DESIGN.md, configured a Supabase database in the background, built the landing page structure, and added sections for the privacy terms and a thank you state. The headline it generated matched the copy file: direct booking and WhatsApp scheduling without commissions.

It did not build multiple separate pages as I initially expected from my prompt. It consolidated the content into a single landing page with the waitlist form. For an initial launch to capture emails, a single page was enough to test the domain.

Testing the waitlist database and reverting a visual tweak

Bolt includes a visual selector tool that lets you click an element in the preview and prompt for changes to just that section. I selected a section and asked it to make the area interactive like a live chart.

The agent made the change, but it did not fit the layout and generated an ugly horizontal scroll bar. Instead of wrestling with CSS prompts, I used the chat to revert the project back to the previous version. That rollback worked cleanly without breaking the rest of the layout.

Then I published the site to a Bolt subdomain to test the form. I filled in the fields with info@mariusbuilds.com, marked the role as a solo barber, and submitted. The record appeared immediately in the integrated database table inside Bolt. When I submitted the same information a second time, the form responded with an error stating I was already on the list.

Connecting GoDaddy DNS and enabling SEO pre-rendering

With the waitlist working on the preview URL, I moved it to the production domain. Inside Bolt’s project settings under Domain and Hosting, I entered salonmate.nl.

Bolt provided the DNS records. I opened GoDaddy, added the required records to the domain settings, and waited for propagation. The connection verified quickly.

In the same settings panel, I turned on SEO Boost. That setting pre-renders the HTML on the server side so search engines like Google can index the content directly rather than having to execute client-side scripts.

Fixing the form validation and checking token usage

After pointing the custom domain, I found an issue in the waitlist form. It allowed visitors to click submit and send data even when required fields were left blank.

I gave the Forge agent a direct instruction: make all the fields mandatory in the waitlist form. It made all the fields mandatory. I tested an empty submit, confirmed the browser stopped it, and clicked update. Bolt pushed the change directly to the live domain.

The entire build, including the initial generation, the rollback, the form fix, and the domain setup, consumed 22 percent of my Forge token allocation. Kimi K3 was one of the heavier models on that plan, which leaves plenty of capacity to build three or four more prototypes on the same allocation.

The design is plain. It relies heavily on text and basic icons, and it needs more graphic work and animation before I would consider it finished. But the waitlist is live, the database is recording sign-ups, and I made in five minutes something I had been postponing for months.

What to copy, even if you never touch Bolt

• Write the copy and the design rules in plain text files before opening an AI editor. The model produces far cleaner layouts when it does not have to invent text and color palettes.
• Test form validation on the published URL, not just inside the preview iframe. AI agents frequently miss basic input validation on the first pass.
• Use a dedicated rollback prompt the moment a visual tweak introduces unwanted scroll bars or layout bugs. Reverting takes seconds; fixing tangled CSS through follow-up prompts wastes tokens.
• Separate your prototype projects from tools with training requirements if your code contains proprietary IP. Forge is cheap because StackBlitz trains on the runs.

Tools mentioned

• Bolt.new: Browser-based full-stack web development environment.
• Bolt Forge: Bolt’s experimental agent tier using open models and reserved infrastructure.
• Supabase: Backend database provider integrated into Bolt projects.
• GoDaddy: Domain registrar used to manage DNS records for salonmate.nl.
• Gemini: LLM used to prepare the landing page copy before building.
• Codex: Used to write the DESIGN.md guidelines file.
• SalonMate: The booking and WhatsApp scheduling landing page built during the test.

Video chapters

• 00:00 Bolt.new before “vibe coding”
• 01:04 Why I’m testing Bolt Forge
• 02:25 The SalonMate idea
• 03:13 Choosing Kimi K3
• 03:45 Adding the copy and DESIGN.md
• 05:09 Forge consent and data use
• 05:42 Checking the token allowance
• 05:59 Bolt builds the website
• 06:47 Free, Pro, and Bolt Lite plans
• 08:43 Reviewing the first version
• 09:54 Testing the waitlist database
• 10:07 Publishing, analytics, and GitHub
• 11:43 Testing the live form
• 12:47 Connecting salonmate.nl
• 13:22 Enabling SEO Boost
• 14:12 How many Forge tokens it used
• 14:39 Fixing the form validation
• 15:13 Final result and verdict