Website

ivy.app is the marketing site for Ivy Tendril. I built most of it: the hero, the header, the software factory section, the download page, and the layer that makes the site readable for search engines and AI agents.

Hero showcase

Discovery. Visitors left the old hero video after a few seconds, and support kept hearing that people did not understand what Tendril does.

Problem. A video is heavy, goes stale with every release, and cannot be read at the visitor's own pace.

Solution. The hero is the app rebuilt in React, playing four scripted stages: plan with chat, refine drafts, review and share, dashboard. It pauses when scrolled out of view, respects reduced motion, and keeps the headline as the largest contentful paint.

Download in the header

Discovery. Analytics showed the download page three clicks deep, with a large drop-off between the home page and the installer.

Problem. The most valuable action on the site was hard to reach.

Solution. The Product menu contains a download panel with tabs for macOS, Windows, and Linux, a primary and a secondary build, and the minimum requirements. The download page repeats the pattern with all three platforms side by side.

Software factory section

Discovery. Sales calls started with the same question: what happens between typing a task and getting a pull request.

Problem. The pipeline was explained in text only.

Solution. One sentence and one diagram: create a plan, agents draft in parallel, drafts wait for review, pull requests come out. Each node shows a count and a tooltip that explains its step.

Introduction video

Discovery. On mobile, the page had no short definition of the product above the fold.

Problem. Visitors had to scroll through feature sections to learn what Tendril is.

Solution. A two-sentence definition followed by a short kinetic type video: open source, multi agent, and the rest of the pitch on a dark background.

Agent readiness

Discovery. An Is Agentic scan of the site scored low, and AI search results described the product incorrectly.

Problem. Crawlers and agents could not read the site reliably.

Solution. robots.txt names each AI crawler, the sitemap carries timestamps, llms.txt and llms-full.txt describe the product, JSON-LD covers the organization and the FAQ, and an OpenAPI spec and agent instructions live at well-known paths. Every control is native, every form field is labeled, and the 404 page returns a real 404. The work covered seventeen audit categories.

Result

Lighthouse SEO 100 and Is Agentic 97 out of 100, with the download one hover away from every page.