Skip to content
AskVault Docs
Try Free →

How to upload PDF and Word documents to AskVault

Written by Joseph Russo, Principal Architect, Networks and Tech, AskVault · Reviewed by Allison Chan
Last updated: · 4 min read

Supported file types

The most common formats (see file uploads for every format by plan):

  • PDF. Most common case. AskVault extracts the text page by page; headings and table layout are not detected.
  • DOCX. Microsoft Word documents. Text in the document body, tables included, is indexed; keeping tables in place with their header row and using headings as section labels is rolling out.
  • TXT. Plain text. No formatting to preserve, fast to ingest.
  • MD (Markdown). Markdown files. Headings, lists, code blocks, tables preserved.
  • CSV. Comma-separated tabular data. By default it is chunked like text; turn on Parse as table to index each row as its own chunk.

Image-only PDFs (scanned documents) need OCR before upload. AskVault does not yet run OCR on scanned PDFs; an image-only PDF indexes no usable text. Run the file through an OCR tool first (Adobe Acrobat's "Recognize Text", Tesseract, or an online tool), then upload the text-searchable result.

Other formats depend on your plan: HTML files from Growth, and spreadsheets and slides from Business. See supported file types by plan. On lower plans, save spreadsheets as CSV and slides as PDF, or crawl web pages with URL crawling.

Upload via the dashboard

  1. Open Knowledge > Add Source > Upload File in AskVault.
  2. Drag and drop your file(s). Or click to browse.
  3. Optional: turn on Parse as table for CSV files.
  4. Click Upload.

Files queue for processing. Larger files take longer to index, and your agent can answer from a file once it's ready.

Progress visible under Knowledge > [filename] > Status.

Uploading through the API

There is no public API for uploading documents today; upload files from the Knowledge page. The public API is for asking your agent questions. See the query API.

How AskVault chunks each format

Format-specific chunking rules:

  • PDF. Text is read page by page in the order the PDF stores it, then split into small passages, at paragraph breaks where the extracted text has them. Headings are not detected, so chunks carry no heading prefix, and tables come through as lines of text without their column layout.
  • DOCX. Body text and tables are indexed and split by paragraph, same passage sizes. Splitting on heading styles (Heading 1, Heading 2) with tables kept in place and their header row on every piece is rolling out and not yet switched on.
  • TXT. Splits on paragraph breaks. Less structural info available; chunks tend to be more uniform.
  • MD. Splits on ## boundaries. Preserves code blocks intact.
  • CSV. Two modes:
    • Text mode (default). The CSV is chunked like ordinary text, so a long row can be split.
    • Parse as table. Each row becomes its own chunk; useful for product catalogues and FAQ tables.

See chunking strategies for the rationale.

Audience after upload

A document's Audience is public (the default) or internal. internal documents are excluded from retrieval on public channels: the widget, the hosted page and the public API.

Set it after the upload finishes, not on the upload call itself:

  • Dashboard. Open the source's settings under Knowledge and pick an Audience.

Bulk upload

For uploading 50+ files at once:

  • Dashboard. Drag and drop several files in one go.
  • ZIP archives. Not supported. AskVault does not unwrap archives; unzip locally and upload the contained documents directly.

For bulk imports from existing knowledge systems (Confluence, Notion), use the native integrations instead of file upload. Those imports are one-time too: re-run them to refresh (see Notion and GitHub).

Plan-level content cap

Total storage per workspace:

  • Free. 5 MB.
  • Starter. 15 MB.
  • Growth. 40 MB.
  • Business. 100 MB.
  • Enterprise. Custom.

Only the extracted text counts toward your allowance, which is usually far smaller than the file itself. Plan upgrades apply straight away.

If you hit the cap, new uploads fail with HTTP 413 Payload Too Large. Delete unused docs or upgrade.

Common pitfalls

PDF text extracts garbled or empty. PDF was created from a scan, so it holds images rather than selectable text. Run it through Adobe Acrobat's "Recognize Text" (or another OCR tool) first, then re-upload the text-searchable version.

Tables look wrong in retrieved chunks. PDF tables are extracted as lines of text, without their column layout. For dense tabular PDFs, convert to CSV first; the CSV ingestion preserves table structure better.

Document indexed but bot doesn't cite it on the widget. Its Audience is internal, so it is excluded from retrieval on public channels. Set it back to public if the bot should cite it there.

Large PDF is refused. Each plan has a per-file limit (5 MB on Free up to 100 MB on Business). Split into smaller files (per chapter or section) and upload separately.

FAQ

Does AskVault keep the original file after indexing?

Yes. The original file stays in storage after indexing, so the document can be re-indexed if indexing has to be retried. Deleting the document deletes the file.

Can I update a document by re-uploading?

Not in place. Every upload creates a new document, even with the same file name, so delete the old document first, then upload the new version.

Does PDF page order matter for retrieval?

Text is read in the order the PDF stores it; there is no reading-order detection, so multi-column layouts (research papers, magazines) can come out with columns mixed.

Can I extract specific PDF page ranges?

Not today. Pre-process the PDF to extract the page range you want, then upload the trimmed file.

What about password-protected PDFs?

Remove the password before uploading. AskVault does not decrypt protected PDFs during ingestion; an encrypted file fails to extract text. Use your PDF tool to save an unprotected copy, then upload that.

Was this page helpful?