Evernote

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 the notes application I actually need instead of Evernote — and start with the export, because the notes are the whole problem.

**Export first, before anything else.** Twenty years of notes inside one product is the lock, and the export is what turns leaving from a decision into a project. Get everything out, check what came with it, and only then decide what to build.

STACK
- Node 20+ with Fastify, server-rendered HTML, or a PWA if you want it offline
- SQLite through better-sqlite3, WAL mode, with FTS5
- tesseract for text in images, run locally
- A browser extension for clipping
- Caddy in front

THE IMPORT, WHICH COMES FIRST
- Parse the ENEX export: notes, notebooks, tags, created and updated dates, and attachments
- Attachments are base64 inside the file and must come out with their filenames and types intact
- The note body is a constrained HTML dialect. Convert to Markdown, keep the original in a column, and report anything that did not convert cleanly rather than dropping it
- Check the counts: notes in, notes out, attachments in, attachments out. An import that silently lost four hundred notes is discovered a year later

THE DATA MODEL
- notes: id, notebook_id, title, body_md, body_html_original, created_at, updated_at, source_url, deleted_at
- notebooks, tags, note_tags — tags cutting across notebooks, because that is how people actually file
- attachments: id, note_id, filename, path, sha256, mime, bytes, width, height, ocr_text
- revisions: id, note_id, body_md, at
- FTS5 over titles, bodies, OCR text and tags

THE CLIPPER
- One click saves the page: the extracted article, the title, the URL and the date
- Extract with a readability algorithm — score blocks by text density and link density, keep the winner, drop navigation and newsletter prompts. Keep headings, lists, quotes, code, tables and figures, because losing a code block makes the clip useless
- **Inline every image and stylesheet into one file.** A clip that still fetches images from the original site rots at the same speed as the site
- A simplified mode and a full-page mode, chosen at clip time
- Selection clipping, which is what people use most

SEARCH, WHICH IS WHY THE ARCHIVE IS WORTH ANYTHING
- FTS5 across everything, instant, including archived notes
- Text recognition over images at import and on new attachments, on device, never sent anywhere. That is what makes a photograph of a receipt findable
- Filters by notebook, tag, date, source and whether it has an attachment
- Record searches that found nothing

WHAT NOT TO CARRY OVER
- Do not reproduce a proprietary note format. Markdown with front matter, one file per note if you want, and the database as an index that can be rebuilt from them
- Write the rebuild command and test it. That property is what makes this safe to trust for the next twenty years

OPERATIONS
- .env: DATABASE_PATH, STORAGE_PATH, BASE_URL, SESSION_SECRET
- Migrations on boot, each once; nightly backup off the machine, restore script
- No analytics. A notes application knows what you are thinking about

WHAT MATTERS MOST
The importer and the counts. Get every note out with its attachments and dates, verify the numbers, and keep the original body alongside the converted one — because the conversion will be wrong somewhere and you will want the source.