PDF Knowledge Base
Open the app: Build a PDF knowledge base → (products.documentize.ai)
In one sentence
PDF Knowledge Base turns a pile of PDFs into a small linked wiki — a summary per document, concept pages that connect them, key facts — and lets you chat with the whole collection, with page citations.
What makes it unique
It is the only app in the book that produces a structure you can browse, grow and download, not just an answer or a single file. Documents can be added later to the same knowledge base (using its ID), concept pages are updated to span old and new documents, and good chat answers can be saved into the wiki. The whole wiki downloads as Markdown files in a .zip.
How it differs from Ask PDF
| Ask PDF | PDF Knowledge Base | |
|---|---|---|
| Purpose | Questions now, in one conversation | A reusable collection you build up |
| Starts with | Files + first question | Files; the question is optional |
| Output | Answers with cited pages | Same answers plus wiki pages and a Markdown .zip |
| Add documents later | No — start a new chat | Yes, to an existing knowledge base |
| Wiki language | – | You can choose it |
| Under the hood | Same search, same citation checking | Same, plus a wiki-writing pass |
When to use it / when not to
- Use it for a project folder, a set of contracts, a course reader, or any collection you will question more than once; or to get a Markdown wiki for Obsidian or a docs site.
- Not for one quick question about one file → Ask PDF is simpler.
- Not for a single polished summary → Abstracts.
Before you start
- PDFs with selectable text; the app's text says up to 10 files and 100 MB per upload. Scanned PDFs show the status "No text (scanned?)" — run Searchable first.
- Decide on a name (for example "Contracts 2026") so you recognise the knowledge base later.
- Keep the knowledge base ID: it is how you come back to or add to it. The page also lists knowledge bases saved in this browser.
Step by step
- Choose Create a new one (give it a name) or Add to an existing one (pick it from your list or paste its ID).
- Drop your PDFs.
- Tick the wiki pages to build and choose the wiki language.
- Optionally type a first question; leave it empty to build the wiki only.
- Press Build the wiki. When done, browse pages on the left (with a filter), read sources, and switch to the Chat tab to ask questions.
- Download wiki (.zip) for the Markdown files; Add documents to extend it; Copy ID to save the reference.
📷 Screenshot placeholder: wiki page list, a summary page, and the chat tab
Options explained
| Option | What it does | Default | Recommended |
|---|---|---|---|
| Summaries | One summary page per document | On | On |
| Concepts | Pages for topics that span documents, linked to each other | On | On for several documents |
| Entities | Shows pages of definitions, examples and other key facts. Key facts are always extracted for the chat; this only decides whether they appear as pages | On | On |
| Explorations | Saves each chat answer as a wiki page | On | On if you will reuse answers |
| Wiki language | The wiki's language; "Same as the documents" by default | Same as documents | Pick one language for a mixed collection |
| First question | Answered right after the build | empty | Optional |
| Answer: Short / Detailed, Cite sources | Same as in Ask PDF | Short, on | Keep citations on |
Summaries and concepts are written in one pass over each document, so switching off either still runs that pass if the other is on; switching both off skips it.
What you get
- A wiki in four sections: summaries, concepts, entities, explorations, linked with wiki links. Statistics show documents, pages, entities, concepts and last update.
- Per document a status: Done, Waiting, Failed, or No text (scanned?).
- A downloadable pdfkb_wiki.zip (Markdown), and cited, highlighted answers in the chat.
- Files and knowledge base are kept about a day (the app's FAQ) and then deleted automatically. Download the .zip if you need it longer. An expired ID shows a message that it is no longer available.
Quality & limits
- AI-generated summaries and concept pages: the page says they "can contain mistakes — check the cited pages".
- Long documents are sampled for the wiki. The summary pass sees at most about 40,000 characters of a document, taken evenly across its pages, so a long book gets a summary of a sample. Chat search still covers every page.
- Time budget. An upload has a limited time for writing wiki pages. Documents that do not fit are deferred and compiled with the next upload; the Write the remaining wiki pages button finishes them.
- Text only. Images and their contents are not analysed.
Works well with
- Use before: Searchable for scans; Merger when many small files belong together.
- Use after: Checklist on a source document for actions; download the wiki into your own notes.
- Recipes: Build a team knowledge base, Digitise a paper archive.
Common mistakes & fixes
- "Status: No text (scanned?)" → OCR that document and upload it again to the same knowledge base.
- "Knowledge base is no longer available" → it expired; rebuild from the originals (and download the .zip next time).
- Lost the ID → check the list of knowledge bases saved in this browser; otherwise it cannot be recovered from the page.
- Summary of a very long document feels thin → it was sampled; ask the chat about specific chapters.
- Mixed-language wiki → set the wiki language explicitly.
Same thing elsewhere
- CLI: a Documentize.PdfKb plugin builds a linked Markdown wiki (a page per document, a concepts glossary, an index) from a folder of PDF, TXT or Markdown files. Its output is simpler than the web version's wiki.
- API: available as an app endpoint. MCP: no dedicated tool found. See channels.
Try it
Upload three short PDFs on related topics as a new knowledge base named "Trial". Build the wiki with all four sections. Open a concept page, follow its links back to the summaries, then ask "Where do these documents disagree?" and check that an exploration page appears. Download the .zip and open one file in a text editor.