Readwise

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me a highlights service that replaces Readwise — and know what makes it ALMOST.

The spaced-repetition part is straightforward. What you are paying for is **imports from everywhere, kept working**: an e-reader, a read-later service, a podcast app, a browser, a notes app, each with its own format and each changing without warning. That maintenance is the recurring cost, and it is the whole verdict.

STACK
- Node 20+ with Fastify, server-rendered HTML
- SQLite through better-sqlite3, WAL mode, with FTS5
- A worker for imports and the daily digest
- A browser extension for web highlights
- Caddy in front

THE DATA MODEL
- sources: id, kind, title, author, cover_path, url, isbn, external_id, imported_from, created_at — a book, an article, a podcast, a paper
- highlights: id, source_id, text, note, location, location_kind, colour, highlighted_at, imported_at, external_id, is_favourite, is_discarded
- tags, highlight_tags
- reviews: id, highlight_id, shown_at, action — the daily digest's record
- imports: id, kind, status, started_at, finished_at, added, updated, skipped, error, raw_path — the raw file kept, so a parsing bug is recoverable
- syncs: id, target_kind, config_encrypted, cursor, last_synced_at — pushing highlights out to a notes application
- Identity is the external identifier where one exists, then a hash of the text plus the source. A re-import must never duplicate, and that rule is the one that decides whether the library stays usable

IMPORTING, WHICH IS THE ONGOING WORK
- One adapter per source, each with the same interface: fetch or parse, produce sources and highlights with stable identifiers, and report what could not be read
- The e-reader's own clippings file is a plain text file with a fragile format — parse it defensively, keep the original, and report unparseable entries rather than dropping them
- A read-later service, a browser extension, a CSV, and an email address that accepts a forwarded highlight
- Every import is idempotent and incremental with a cursor
- Keep the raw file for every import. When an adapter turns out to have been subtly wrong for six months, that archive is the only way back
- Each adapter has a fixture and a test. They break when the source changes, and a failing test beats a mystery
- Say in the README that this is the recurring cost, and build only the two or three you actually use

THE DAILY REVIEW, WHICH IS THE POINT
- A small number of highlights each day, by email and in the interface. Five is enough; fifteen is a chore
- Selection weighted towards things not seen recently, favourites, and recent additions, with everything eventually resurfacing
- Actions on each: favourite, discard, tag, add a note, or open the source
- Discarding must be real and permanent for that highlight, or the digest fills with things you have already decided you do not want
- The email is plain and readable, with a one-click way out. That daily mail is the entire habit, so it must be worth opening

SEARCH AND EXPORT
- FTS5 across highlight text, notes, titles and authors, with filters by source, tag, date and favourite
- Export everything as Markdown grouped by source, as CSV, and as JSON, in one command
- Push to a notes application on a schedule: one file per source, appended rather than rewritten, with a stable identifier per highlight so an edited note is not clobbered. That sync is what most people actually want and it is where a naive implementation destroys somebody's annotations

THE EXTENSION
- Select text on a page, highlight it, add a note. The article's title, author and URL captured with it
- Talks only to your own server, per-host permissions rather than every site

OPERATIONS
- .env: DATABASE_PATH, STORAGE_PATH, BASE_URL, SESSION_SECRET, SMTP_URL, ENCRYPTION_KEY
- Migrations on boot, each once; nightly backup off the machine — this is years of reading
- Health endpoint reporting the age of each import source

WHAT MATTERS MOST
Stable identity across re-imports, and the daily email. The first keeps the library clean; the second is the only reason any of this is worth having.