EN
Keep me posted
All insights

Case

How zaela.com was made with Zaela

An honest case: the brief, how the work was split between memory, agents, skills and tools, and what a person still decided.

Zaela, made by Edge4 min read

zaela.com is the international site of Zaela, the new name of Edge OS. It speaks five languages: English, French, German, Dutch and Simplified Chinese. It was built by Claude, working inside Zaela. This case shows how the work was divided, and where a person kept the last word.

The brief

Introduce Zaela, the new name of Edge OS, to an international audience, in five languages, with a brand film at the centre. The site is also its own proof: it was made with the product it describes.

Memory: the brand, written down

The brand memory guided the work: the brand rules, the tone of voice, the colours and the house style. Those rules are what keep a site in five languages recognisably one site.

The house style is precise enough to follow. The logo is a butterfly made of 88 dots on a hexagonal grid. The whole site speaks the same “dot language”: animations, pictograms and fields of dots, which never sit behind text. Rules this exact are what memory is for: written down once, they guided every page.

Agent: a written method

The second guide was the method of the Website Builder agent. It is written down and shown in full on the site’s Website Builder page, in seven steps: Goal and market · Promise and structure · Style frame · Stack · Build · Test · Deliver and go live.

On Zaela, a method can live in a separate skill, shared by several roles, or inside one agent’s own instructions. Here it was the second: the method belongs to the Website Builder.

The technical choices follow the same line. The site is static: a small build script turns templates and five language files into plain HTML pages. There is no framework. That matches the method’s own advice: static where it can be, with the core content as real HTML that search engines and AI crawlers can read.

Tools: the film and the publishing

The brand film is a natural-history film of 1 minute 29 seconds about a real butterfly, Dismorphia zaela, known in English as the zaela mimic white. It has two chapters: I · Mimicry and II · The butterfly effect.

Each part was made in its own way:

  • Footage: generated with AI, using Google Veo 3.1 Fast, from keyframes made with an image model. The footage is labelled as AI-generated.
  • Voice-over: synthetic, made with Gemini text-to-speech.
  • Music and sound: synthesised in code, in F sharp major.
  • Subtitles: in all five languages.

Some context on the models. Google introduced Veo 3.1 and Veo 3.1 Fast in October 2025. Veo 3.1 can work from images you give it, such as a first and a last frame, so a still image can fix how a shot looks before any motion is generated. Gemini’s text-to-speech lets you steer style, accent, pace and tone in plain language.

Publishing runs through a tool as well: Zaela’s own web publishing. Every version goes to a preview first. Going live needs approval from an Owner or Admin.

Checks before every release

Before every release, each page is checked automatically in all five languages, on desktop and on mobile, for errors and for layout that overflows.

What a person still decided

Claude did the building, guided by memory and method. A person made the choices of style and the choices for the film. And no version goes live without approval from an Owner or Admin.

That is a useful split. The AI does the work that can be done well and repeated; people decide where taste and responsibility come in.

What you can take from it

  • Write the brand down first. An AI cannot follow brand rules it has never been given. Precise rules, like “dots never sit behind text”, carry across languages and pages.
  • Give the work a method, not a template. A written method with a quality bar travels to the next project; a template only fits the last one.
  • Choose the simplest stack that works. Plain HTML from a small build script is easy to check.
  • Label what AI made. The film’s footage is AI-generated and labelled as such. Say which tool made what, as this case does.
  • Automate the checks. Five languages times two screen sizes makes ten views of every page. Let a script look at all of them before every release.
  • Keep a person at the gate. A preview first, then approval to go live.

Your own version

On Zaela, your team works with the same four building blocks: a brand memory, agents with a written method, tools that make and publish, and approval where it counts. See what Zaela does.

How would Zaela work for you?

Answer six short questions and get a first sketch, with the agents that fit.

Try the advisor

Read next

Stay informed

Curious what Zaela could become for you?

Leave your details and we’ll get in touch to tell you more about Zaela.

We use your details only to contact you about Zaela.