kindle

A serverless read-later for the Kindle experimental browser. The repo is the database, GitHub Pages is the host: you add text, a static site rebuilds, and it's live at neves.cloud/kindle in ~1 min. Nothing runs — the push-to-live latency is the "real-time" here.

Built mostly for Claude-generated text — a long answer, a synthesis, a draft — piped straight onto the device to read on e-ink instead of in a terminal. Saving a web article works too.

Add something to read

… | bin/kindle-add --stdin --title "On focus"   # pipe markdown in (the main path)
bin/kindle-add --md notes.md                      # a local markdown file
bin/kindle-add https://example.com/article        # fetch + extract a web page
bin/kindle-add <…> --no-push                       # build locally, don't publish

Text forms take markdown directly; a leading # H1 becomes the title (else --title, else the filename) and is dropped from the body so it isn't repeated. The URL form uses trafilatura to extract the article to markdown. All three write docs/a/<slug>.json, rebuild, commit, and push. Reload on the Kindle to see it.

How it renders

build.py turns each docs/a/<slug>.json into a clean e-ink page and regenerates docs/index.html (the reading list). No runtime JS or fetch — plain static HTML the Kindle's ~2012 WebKit just renders.

Constraints (why it's built this way)

• Static only — no server; the repo is the store, Pages serves it.
• ~2012 WebKit — server-rendered static HTML, no flex/grid.
• E-ink — black/white, large serif, block layout, justified paragraphs.
• Subpath — served under /kindle/, so asset links stay relative.

docs/index.html and docs/a/*.html are generated — never hand-edit; run build.py.