 |
 |
 |
 |
Turn any GitHub repository's history into an animated film you can watch, scrub, and share in the browser.
Paste a repo URL, get back its whole life: directories growing, contributors arriving and leaving, releases landing, dead code rotting away. That's the goal. The name is a working title.
Status: early development. The rendering experiments work on real data from a handful of bundled repos. The pipeline that ingests an arbitrary GitHub URL comes next, so this is not usable as a product yet. Watch the repo if you want to follow along.
Before committing to one visual style, I'm building several competing ones as standalone "sketches". Each is a full renderer over the same event stream, with a shared chrome: sidebar with live stats and a contributor leaderboard, date display, and a timeline you can scrub. Playback supports pause, fast forward, and jumping anywhere in history. There is a dropdown to switch between the bundled repos (vite, react, svelte, express, chrono).
| Sketch | The repo as... |
|---|---|
/sketches/civilization |
a top-down empire. Folders are cities, commits raise rooftops, merges send caravans, the top contributor wears a crown |
/sketches/city |
a skyline in time-lapse, with construction cranes for branches and a day/night cycle per repo-year |
/sketches/gource |
a force-directed file tree, after Andrew Caudwell's Gource |
/sketches/race |
a contributor bar chart race over a directory streamgraph. No metaphors, just charts |
/sketches/planet |
a living world whose territories are directories |
/sketches/layout-test |
a 10k-node force layout stress test |
Most things respond to hovering and clicking, and a tuning panel exposes each sketch's parameters.
Ingestion is designed to happen once per repo version, not per viewer. A
worker does a bare, blob-less clone (git clone --bare --filter=blob:none)
and walks git log --raw, which compares tree entries by hash and never
downloads file contents. For vitejs/vite that means 9,308 commits fetched in
about 30 seconds and 16 MB, compiled into a 1.3 MB JSON artifact.
The artifact is the contract between the data side and the visual side: sketches and films only ever consume events (commit, merge, release, mass delete, new contributor), never raw git data. Artifacts get cached in object storage and served with zero egress fees, so a repo going viral costs about as much as a popular static file.
Rendering happens entirely client-side with PixiJS v8, targeting 60fps on mid-range hardware. WebGPU where available, WebGL otherwise, and reduced-motion preferences are respected.
You need Node 22+ and pnpm (the exact version is pinned in package.json).
git clone <this repo>
cd repomentary
pnpm install
pnpm dev # http://localhost:3000Other commands:
pnpm check # lint + format (Biome)
pnpm typecheck
pnpm build
pnpm preview # production build in the Workers runtime
pnpm deploy # deploy to Cloudflare WorkersTo extract a new repo's history for the sketches:
git clone --bare --filter=blob:none https://github.com/OWNER/NAME.git /tmp/name.git
node apps/worker/scripts/extract-history.mjs /tmp/name.git OWNER/NAME apps/web/public/data/name.jsonthen register it in apps/web/src/lib/realHistory.ts.
| Path | Contents |
|---|---|
apps/web |
Next.js app: site, sketches, and eventually the player |
apps/worker |
ingestion worker (placeholder) and the extraction script |
packages/artifact |
event/artifact types plus a synthetic history generator |
- Foundation and visual identity (where things are now)
- Ingestion pipeline for arbitrary GitHub URLs
- The film itself: the sketches grown into selectable film styles
- Sharing and launch
- Accounts and saved repos
- Embeds and themes
- Video export
Everything is still in flux, so large PRs will probably collide with ongoing rewrites. Issues and ideas are welcome. If you want to build something bigger, open an issue first so we can talk.
The gource sketch is a from-scratch homage to Gource by Andrew Caudwell. Bundled sample histories come from the public repos of vite, react, svelte, express, and chrono.