Through running a conversion rate optimization agency, I am constantly sifting through web pages to find small attributes that could lead to higher conversion. I will often compare many pages at once from different companies in a given niche, to understand the similarities and differences I can use to experiment with. But that process is quite slow: go to a webpage, use a Chrome extension like GoFullPage to capture the screen, then copy and paste into a Figma file one by one. And those captures go stale as competitors change their websites over time, making you repeat the process again and again.
With Haya, you can simply ask in natural language for what you are looking for, like pricing pages from Notion and its competitors, and it will surface those pages for you to compare or export to Figma, using computer-use agents to visit each site, capture screenshots, and return them. Always live, always on. You can even give it context on you and your company, so you can ask for things like the top landing pages of my competitors.
- Indexing websites quickly by going through their XML sitemaps, combined with crawling and AI URL resolution, to turn a plain-English request into the exact set of pages to capture.
- Building a Playwright capture engine that produces reliable full-page screenshots: intelligent scrolling, lazy-load handling, and batching many sites with configurable concurrency.
- Designing a pluggable storage layer so captures can land on the local filesystem, AWS S3, or Supabase Storage, with a manifest written for every run.
Haya is split into a TypeScript backend on Railway, exposing both an Express HTTP API and a CLI for discovery, capture, and storage, and a Next.js frontend on Vercel with the AI chat interface, real-time progress tracking, and an image gallery, with authentication handled by Supabase.