ShogunAI

The brief: a product people need to see working

ShogunAI is a Mac app that keeps a private memory of your work, your mail, meetings, documents and what was on your screen, so you can ask about any of it and get help with whatever comes next. It launched publicly in October 2026. We came to it through the product: we had already redesigned the app’s dark interface, then designed and built the second version of its launch site.

Every AI assistant now claims to save time, and the claim costs nothing to make. What sets this one apart, that it starts from your week instead of from a blank prompt, is only believable when you watch it answer a real email with context it could only have from memory. So the brief we wrote for ourselves was short: show it doing the work, and keep everything else out of the way.

One real task in the first screen

The hero carries the headline, two actions and a laptop. The actions are the two decisions a visitor can make on arrival, try the app or watch the launch film, and the free trial sits directly under them so nobody has to scroll to find the terms.

The laptop is not an illustration. It shows the app drafting a reply to a message about Friday’s launch, inside Gmail, with the context pulled from earlier in the week. Mount Fuji sits behind it as the brand’s setting, but the eye goes to the screen, because that is where the proof is.

Motion that was storyboarded before it was built

The page tells several short stories: how the memory fills up through a week, what the app does when you take a closer look, how a sentence appears at the caret when you press Option. Each one was storyboarded before any code was written, frame by frame, so the client could approve the story rather than react to an animation.

In the build, the stories are scroll linked. A section holds still while its story plays, then lets the page move on. Smooth scrolling and scroll timing run on Lenis and GSAP’s ScrollTrigger, and every other section uses one quiet reveal, so the stories stand out instead of competing with decoration.

Answering the price question before it is asked

People reaching the lower half of a launch page are working out whether it is worth paying for. The page meets that directly. One section lists the separate tools the app replaces, meeting notes, translation and reply drafting, each with its own monthly price, against one subscription. The productivity chart beside it says plainly that it is an illustrative scenario, not a measured result, which is the kind of honesty a sceptical buyer notices.

The plans follow: two of them, a monthly and annual toggle with the saving marked, and the free trial repeated above the cards. The closing section turns the annual price into about $1.61 a day.

Privacy, said where it matters

A product that remembers everything has one question to answer before any other: where does that memory live. The page answers it in its own section, your memory stays on your PC, rather than in a footnote.

The site follows the same rule. Analytics are optional and off until a visitor allows them, and the prompt says exactly what is measured and that no email or form content is sent.

Built to be handed over

The front end is a static page with no framework and no build step on deploy. It is assembled from 14 section files, with design tokens mirrored from the Figma file and one script per section, so the client’s engineers can change a section without touching the rest. Mac visitors see Download for Mac and Windows visitors see Join the Windows waitlist, decided before the first paint so the wrong button never flashes. Every video carries a poster image, so the page reads correctly before a single clip loads. The code was handed over as a repository the team now owns.