kral Documentation
Sign in
Tools

Files

Drag a file into the prompt box and ask. That is the whole workflow: no pipeline, no preprocessing.

What you can upload

  • PDFs: text PDFs are read directly; scanned PDFs go through OCR (see below).
  • Word and Excel: documents keep their structure, tables are read as data.
  • Images and screenshots: vision-capable models describe, read, and reason about them.
  • Source code: keeps its structure; upload single files or archives.

Several files can go into one message, within the per-upload file limit and size limit shown in the chat.

OCR for scans

Scanned documents and photos of documents are run through a dedicated OCR model that handles layouts, tables, and handwriting far better than plain text extraction. OCR is billed per page and the cost shows under the answer like everything else.

With the file search badge enabled, the model searches across your uploaded files instead of reading them linearly. This is the right mode for large document sets: it embeds the content once and retrieves the relevant passages per question, which is faster and much cheaper than re-reading everything per prompt.

Where files live

Uploaded files belong to your account and stay attached to their conversations. Delete a file or its conversation, and it is gone. Files attached to agents as knowledge stay until you remove them from the agent.