Outline

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 the team wiki I actually need instead of a hosted Outline: collections, a fast editor, search that stays fast, and single sign-on.

Read this first: Outline is open source and self-hosting it is free — the fee buys a managed instance with backups, upgrades and single sign-on already configured. Build this when you want it to be exactly yours; otherwise self-host the original, which is very good.

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 DATA MODEL
- collections: id, name, slug, description, icon, colour, permission_default, position
- documents: id, collection_id, parent_id, url_id, title, body_md, body_html, emoji, created_by, updated_by, published_at, archived_at, deleted_at, created_at, updated_at
- revisions: id, document_id, title, body_md, author_id, at — every save, so nothing is lost
- backlinks: from_document_id, to_document_id — resolved at save time
- memberships and permissions: per collection and, where needed, per document — read, write, admin
- shares: id, document_id, token, include_children, password_hash, expires_at, revoked_at, views
- stars, views, comments, and a search_index over titles and bodies
- The url_id is a short stable identifier in the URL alongside the slug, so a renamed document keeps every link that has ever been shared. That one decision saves more broken links than anything else here

THE EDITOR
- Markdown underneath, rendered as you type. The file is always Markdown, which is what makes an export trustworthy
- Slash commands for headings, lists, checklists, code, quotes, tables, images, dividers, and embeds
- Nested documents, so a collection has structure without a folder tree
- Autosave, with no save button and no draft limbo
- Paste from a document or a web page keeps structure and drops styling
- Typing a bracketed title links to another document and offers to create it

COLLABORATION
- Real-time editing with presence over WebSockets, using an existing conflict-free replicated data type library for the text. Writing one is a research project and getting it subtly wrong loses people's work
- Comments anchored to a selection, threaded, resolvable, with a notification to the people mentioned
- If real-time is too much, a per-document lock with a clear indicator is the honest fallback — worse, never destructive. Choose deliberately and write down which

SEARCH, WHICH DECIDES WHETHER THE WIKI IS USED
- FTS5 over titles and bodies, with titles ranked far above bodies
- Filters by collection, author, date and document status
- A keyboard shortcut from anywhere, arrows to move, Enter to open. This becomes the main navigation
- Search must stay fast as the workspace grows: index at write time, never scan at read time, and keep the index in the same database so a search is one query
- Record searches that return nothing — that list is the set of documents somebody expected to exist

PERMISSIONS
- Per collection by default: private, read for the workspace, or write for the workspace
- Per document only where genuinely needed. Fine-grained permissions everywhere is how a wiki becomes a place nobody can find anything
- Checked on the server in the query, default deny, including in search results — filtering results after the query is how a private title leaks
- Guests limited to named collections

SHARING OUTSIDE
- A public link with a long token, optionally including child documents, with a password and an expiry, revocable
- A shared document is server-rendered, fast, and exposes nothing but itself
- Views counted per share

HISTORY AND SAFETY
- Every revision kept, with a diff and a one-click restore that writes a new revision
- Archive rather than delete, with a trash and a stated retention before anything actually goes
- An undelete that puts the document back where it was

INTEGRATIONS
- A search endpoint your chat tool can call, which is where half the searches will come from
- Webhooks on document published and updated
- Import from Markdown, from HTML, and from the common exporters

EXPORT
- The whole workspace as a directory of Markdown with attachments and structure intact, in one command
- Any document as Markdown, HTML or PDF
- The README says this plainly, because being able to leave is what makes it safe to arrive

ACCESS
- Single sign-on through OIDC, which is what the paid tier charges for and is a few hours' work
- Magic links as a fallback, sessions revocable centrally

OPERATIONS
- .env: DATABASE_PATH, STORAGE_PATH, BASE_URL, SESSION_SECRET, OIDC_*, SMTP_URL
- Migrations on boot, each once
- Nightly backup off the machine, restore script — this holds the team's institutional memory
- Health endpoint

WHAT MATTERS MOST
Search and permission checks inside the query. Build the index at write time and make sure a private document cannot appear in anybody's results, then use the search as your only navigation for a week. A wiki is only as good as the ability to find the page you half remember.

Give me the repository, migrations, .env.example, a seed workspace, and a README with deploy steps behind Caddy.