Import PDFs, text and images

Five source types with different validations

Wavize accepts a document, image, web page, pasted text or full website. Supported document formats are PDF, DOCX, TXT and MD; images may be PNG, JPG, JPEG or WEBP. Maximum size and some image functions depend on the plan.

Importing a file does not automatically make all its content reliable. Structure, text quality, date and usage authorization remain essential. A blurry scan or context-free table may produce insufficient extraction.

Confidentiality
Remove passwords, API keys, banking data, medical files and unnecessary personal data before import. Only information the assistant may legitimately use should enter its knowledge.

Prepare a readable, versioned document

A few minutes of preparation prevents contradictions that are difficult to diagnose.

1. Choose the right format

   Prefer PDF or DOCX for structured layout, TXT or MD for simple text. Do not merely rename an extension: actual file content must match its type.

2. Make text selectable

   A PDF containing real text is more reliable than a series of images. For a scan, perform clean OCR and review numbers, dates, product references and currency symbols.

3. Mark version and scope

   Include date, market or range in both the document and Wavize title. Archive or deactivate the old source when replacing it.

Import according to available content

Choose the matching type: the form does not expect the same fields for a file, image or text.

1. Document or image

   Enter a title and select the file. The server checks format and size. Image indexing may be refused if the plan does not include that feature.

2. Pasted text

   Choose Text, enter a title and paste structured content with complete sentences. This mode suits a short FAQ or a procedure you own.

3. Start training

   The option is enabled by default. The file is saved and processed in the background; do not delete it merely because status is not immediately complete.

Check what indexing actually understood

Test areas most likely to degrade: tables, footnotes, images and short references.

• Atomic questions

  Ask one fact per question: duration, price, condition, exception. An overly broad question does not reveal which document part is problematic.

• Tables converted to sentences

  If rows are mixed, add a text version: “For plan A, the limit is X.” An explicit caption also improves column understanding.

• Images as supplementary source

  Do not entrust a critical rule solely to a screenshot. Add the same information as text so it remains accessible, searchable and easy to update.

Criteria for a usable document

A successful import should satisfy these functional checks.

• Accepted file — Format and size comply with the plan without validation errors.
• Processed source — Training status is complete and the source remains active.
• Sensitive facts reviewed — Dates, amounts, exceptions and contact details match the original document.