One feed. Every source.

Sources

• Videos

  @MatSiems + @mat-siems-production on YouTube.

• GitHub

  Stars from @flexappdev.

• Prompts

  Top 100 AI prompts.

• Apps

  Fleet apps catalogue.

• Sites

  Curated sites worth scrolling.

• Wikipedia

  Cached Wikipedia articles — full copy, sections, media.

• WikiVoyage

  Random WikiVoyage destinations.

• Amazon

  Amazon UK Best-Sellers (zgbs).

• Images

  S3 image gallery with metadata + search.

Stack

• Next.js 15
  App Router · React 19 · Server Components
• Tailwind CSS
  Dark theme · zinc/emerald tokens
• Supabase
  Auth + Postgres CMS (scroller_sites)
• MongoDB
  AmazonDB.topSellers + AzDB.MAIN for Amazon picks
• AWS S3
  com27 bucket with signed URLs for the image source
• Vercel
  Hosting + edge caching · deployed to scroller-psi + scroller-bay

Architecture

Nine sources are fetched by Next.js server components, normalised through a shared fetcher layer, and rendered by a single PageBrowser client component that drives scroller, list, and grid views across every page. On desktop (lg+) clicking a card opens an inline side-panel instead of a modal overlay, keeping the grid visible.

Light / dark toggle in frame · full screen: open diagram

Cache & fallback flow

Every source has a cache layer — unstable_cache for live APIs, MongoDB for the 67k WikiVoyage corpus, S3 signed URLs for images — so a single cold lambda serves warm data within 10 minutes of a cache miss.

Dashed lines = fallback path · full screen: open diagram

Content generation pipeline

Every image, video, and diagram that ships with Scroller is generated by an ABC skill or one of the in-repo scripts and uploaded to s3://com27/scroller/ — idempotent, re-runnable, and version-pinned to Runware FLUX (images) + Seedance (video).

Accent lines = AI-generated assets · full screen: open diagram

Attribution & disclosures

Part of the fleet

Scroller is one of 15+ sites Mat Siems ships under the flexappdev and cleverfox-ai GitHub orgs. The full catalogue is browsable in /apps.