PD AI Assistant can extract text from supported documents and index it as custom knowledge. It answers from extracted passages rather than preserving a document’s page design, images or interactive elements.

Prepare a readable document

The upload control accepts .txt, .md, .pdf and .docx, with an indicated maximum file size of 8 MB. Use the actual supported file format; renaming an unsupported file does not convert it.

For PDF, prefer selectable text. The module does not provide OCR for scanned images, so an image-only PDF may yield no usable text. An encrypted or damaged PDF can fail extraction. A Word upload uses the document’s text; do not assume information held only in an image or a complex visual layout will become a reliable answer source.

Upload and index

1. Open PD Modules → PD AI Assistant → Settings → Knowledge.
2. Check that an indexing provider is available.
3. Under Add a source, choose the intended audience in Who can this reach?.
4. Use Upload a file (optional) to select the document.
5. Review the title filled from the filename and make it useful for later citations.
6. Choose Add to knowledge base, then check the result and Indexed sources.

Text and Markdown files populate the text box for review. PDF and Word text are extracted when you add the document; selecting the file is not an extraction preview. Do not assume that a previewed paste has replaced an already selected binary document.

Check the result, not just the upload

Verify the title, visibility, passage count and embedding provider. Ask a controlled question about an important statement in the document and compare the answer with the original. Check numbers, conditions and steps that may depend on layout.

The document text goes to the embedding provider during indexing. Choose material appropriate for that provider and audience before uploading; selecting Staff only controls retrieval, not whether indexing sends text off the server.

If extraction fails, use a readable text-based document or paste reviewed plain text. For a missing server capability, retain the redacted error and have your server administrator check the supported installation requirements; do not install libraries or run guessed repairs to hide it. Check the source list before retrying an uncertain add.

See the PD AI Assistant Blesta guide for related tasks, or explore PD AI Assistant.