BookStack Cloud

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 documentation wiki in the shape BookStack has — and note that the software is open source and free, so build this when you want the model to be yours.

STACK
- Node 20+ with Fastify, server-rendered HTML with a little vanilla JS
- SQLite through better-sqlite3, WAL mode, with FTS5
- Caddy in front

THE STRUCTURE, WHICH IS THE IDEA
- Shelves hold books, books hold chapters, chapters hold pages. Four levels, fixed, and no configuration
- That is a physical metaphor everybody already understands, and it is why people find things without being shown. Do not make it configurable — the constraint is the feature
- A page can live directly in a book without a chapter, because forcing a chapter for one page is the kind of ceremony that stops people writing

THE DATA MODEL
- shelves: id, name, slug, description, position
- books: id, shelf_ids_json, name, slug, description, cover_path, position, archived_at
- chapters: id, book_id, name, slug, description, position
- pages: id, book_id, chapter_id, name, slug, body_md, body_html, draft_of_id, position, created_by, updated_by, published_at, archived_at
- revisions: id, page_id, name, body_md, author_id, at, message — every save, with a diff and a restore
- links: from_page_id, to_page_id, resolved at save
- attachments, comments, watchers, tags
- permissions: subject_kind, subject_id, scope_kind, scope_id, action, effect — inherited down the hierarchy, with an explicit rule overriding
- Slugs are permanent. A renamed page keeps its slug and writes a redirect rather than breaking every link anybody has shared

WRITING
- Markdown with a live editor, and a WYSIWYG alternative for people who will not use Markdown. Both must produce the same stored Markdown
- Slash commands for the blocks that matter, and Markdown shortcuts while typing
- Drafts autosaved separately from the published page, so an unfinished edit is never what a reader sees
- Templates per book, for the document kinds a team writes repeatedly
- Attachments stored by hash with the type from content, served through a permission check

PERMISSIONS
- Inherited: a page from its chapter, a chapter from its book, a book from its shelf, with an explicit rule overriding
- **Deny beats allow at every level.** Getting the precedence wrong once means a confidential page visible to everyone, and nobody discovers it deliberately
- Resolved into an effective permission per person per page and cached, invalidated when any rule in the chain changes
- Checked inside every query including search. Filtering results afterwards is how a private title leaks into an autocomplete

SEARCH
- FTS5 over titles, page bodies, and the extracted text of attachments, with titles ranked far above bodies
- Filters by shelf, book, tag, author and date
- A keyboard shortcut from anywhere; this becomes the main navigation
- Record searches that found nothing — that list is the pages somebody expected to exist

STAYING FRESH
- An owner and an optional review period per page. When it lapses the page says so at the top
- A wiki dies by filling with pages that were true once, and this is the cheapest defence there is
- Archive rather than delete, with a restore

OPERATIONS
- .env: DATABASE_PATH, STORAGE_PATH, BASE_URL, SESSION_SECRET, OIDC_*
- Migrations on boot, each once; nightly backup off the machine, restore script
- Export the whole workspace as Markdown with the structure and attachments intact, in one command
- Health endpoint

WHAT MATTERS MOST
Effective permissions and the fixed hierarchy. Cache the permission resolution and apply it inside the query, and resist every request to make the structure configurable — the four levels are why people can find anything.