Ingesting sources

Ingestion is how a source becomes part of your wiki. Add something; Wikai reads it and writes linked pages.

How to add a source

• Drag and drop a file onto the wiki — a drop overlay appears.
• Add a source card on the wiki landing page (“drop a file here or click to browse”), or the small + in the toolbar.
• Paste a URL into the source card and click Fetch to ingest a web page.

Supported formats

Files Wikai can’t read

A file with no importer — a PowerPoint .pptx, say — is stored, not ingested. Wikai keeps the original under raw/, writes stored to the wiki’s log, and shows an amber row in the import panel naming the file type that has no importer. Nothing is silently dropped, and nothing pretends to have worked.

PowerPoint is the notable case: .pptx has no importer yet. To work with a presentation’s content now, export it to PDF in your presentation app and import the PDF.

The same amber row appears for a file Wikai recognizes but won’t read. A DRM-protected EPUB — Adobe ADEPT, Apple FairPlay, Readium LCP — is stored with a note naming the scheme. Wikai does not remove DRM; DRM-free and watermarked EPUBs import normally.

What happens

Your file is copied into the wiki’s raw/ folder (untouched), processed, and written out as one or more pages under pages/. A line is added to log.md, and the page list and graph refresh automatically. Larger PDFs show a per-page progress panel while they extract.

If the machine runs short of memory while a document is being read, the same panel says so — since which page, and that those pages will be slower — and clears the note when memory recovers. The slowness is the machine’s; Wikai says so rather than looking hung.

Books

Drop an .epub on a wiki and it becomes a book: a book index page and a chapter page for every section in the book’s own table of contents. A volume that packs several books into one file becomes one page per book, and a section that straddles the publisher’s file breaks becomes one page. The title, authors, language, date, subjects and identifiers come from the book’s metadata.

Illustrations are kept at full size — Project Gutenberg’s habit of showing a thumbnail that links to the big picture is undone, since the link had nowhere to go inside a wiki. Every paragraph keeps its anchor, so links into a chapter — from the book’s own contents page, from your notes — resolve after the import.

Chapters arrive locked: the text stays faithful to what you ingested. They are exactly the pages comments and highlights are for.

PDFs in detail

A PDF doesn’t store a document — it stores marks on a page. Getting a readable Markdown page back out is reconstruction, and it’s where most of Wikai’s ingestion work goes.

• Columns are read in the right order, rather than stitched straight across the gap between them.
• Tables become Markdown tables. When a grid genuinely can’t be reconstructed, the table is captured as an image instead of collapsing into a column of loose numbers — a picture of a table beats a scramble of one.
• Mathematics survives. Display equations become LaTeX and render; inline math stays inline.
• Footnotes become real footnotes — a reference where it belongs, the definition kept at the end, rather than the note being spliced into the middle of the sentence that called it.
• Links inside the PDF are kept as links.
• Figures are captured, including ones drawn as vector graphics rather than embedded as images.
• Headings are told apart from emphasized body text, so the chapter outline means something and numbers survive intact. Numbered headings land at the right depth, bold labels stay bold text instead of becoming headings, and bullets and numbered items become lists.
• Chapters, with or without an outline. A PDF with bookmarks is split into chapters on them. A PDF of twenty pages or more with no bookmarks but numbered headings is split on those headings, so it reads the same way. The page summary is drawn from the first real prose, not from the table of contents.

Scanned PDFs

A PDF with no text layer — a photographed book, a scanned report — used to arrive as a placeholder saying so. It’s now read with OCR and becomes searchable text like any other source.

• Per page, not per document. A born-digital report with scanned appendices pays for the appendices only.
• No prompt. OCR starts when a page needs it. The ingest panel shows progress and an estimate, and you can cancel from there.
• It runs on your machine. Nothing is uploaded — the page never leaves your disk. The recognition models are downloaded once, the first time you ingest something that needs them, checked against a known checksum, and kept on your machine afterwards; the ingest panel shows the fetch. After that, scanning works offline.
• Figures are cropped out, not the whole page. A scanned page that is mostly text arrives as text, with its figures and charts cut out and placed where they belong — like a born-digital page. A page that is mostly picture keeps the full-page image. Scanned tables are kept as crops rather than guessed at.
• Latin script. The bundled recognition model covers Latin-script languages. Other scripts aren’t read yet.
• Every page says whether it was OCR’d. The metadata panel shows ocr and which engine produced it, so a recognized page is never mistaken for a faithful extraction.

If the OCR engine can’t be loaded, that’s a warning rather than a failed ingest: the page is still written, with an honest placeholder instead of invented text.

Word documents and the pandoc family

.docx, .odt, .rtf and LaTeX go through pandoc. The images in them appear on the page, and the HTML fragments pandoc leaves behind for a table or between two lists are rewritten as Markdown where that loses nothing, so the editor can show them. Long documents convert as quickly as short ones; the “timed out” failure that used to hit anything past a few pages is gone.

If a file already exists

When you add a file with a name that’s already there, Wikai offers to rename (the default — appends a number), overwrite, or skip.

After ingesting

Open the new pages, search across them, ask the chat sidebar a question grounded in what you added — or read a chapter and comment in its margin.

Last updated 2026-10-07.