Tavina

Uploading files and scans

๐Ÿ“– This content is synced from the Tavina Handbook.
The canonical version always lives at https://tavina.ai/docs/friendly/files-knowledge/uploading-files

Uploading files and scans

You can upload files so Tavina can use them as context. Document-like uploads are part of the Knowledge Library foundation: they can become Sources, land in a Catalog, and be processed into extracted text and chunks for retrieval.

How to upload in the PWA

  1. Open Knowledge Library from the toolbar.
  2. Choose where the file should go:
  • use Library Inbox if you are sorting later;
  • choose a personal Catalog if the file already belongs somewhere specific.
  1. Use the add/upload Source control in the Knowledge Library panel.
  2. Choose the Source type Tavina asks for, such as a file upload, URL, pasted text, Markdown, or another supported Source type.
  3. For a file upload, select the file from your device. If your browser supports drag-and-drop in the active upload area, you may use that; otherwise use the file picker.
  4. Give the Source a title that will make sense later.
  5. Wait for processing to finish before treating answers as complete.

After upload, keep the Knowledge Library panel open long enough to confirm the new Source appears in the expected Catalog and has a usable processing state.

Chat attachments in a single message

Chat attachments are for context you want Tavina to use with the message you are sending now. They are separate from uploading a Source into the Knowledge Library, although document-like files may still be handled in a Source-aware way behind the scenes.

You can attach up to 20 files to one chat message. Tavina should carry every accepted attachment with the prompt instead of only using the first file. If you try to add more than the limit, Tavina should show a clear message and should not quietly upload, send, or drop the extra files.

Image attachments appear as separate thumbnails in the conversation. Select a thumbnail to open a full-size preview for that image. Documents and other non-image files appear as privacy-safe file cards or chips, with compact details such as file type or size rather than private storage paths or full extracted text.

If a security scan, validation step, or upload fails for one file, that failed file should stay out of Tavina’s context. The conversation should still show the upload or scan error so you can remove it or try a safer copy.

What the upload UI tells you

The exact wording may vary by Source type, but the important signals are:

  • the Source title;
  • the Catalog where it landed;
  • the Source type;
  • the import or processing status;
  • any error summary, partial extraction note, OCR-needed state, or blocked/encrypted state.

If the new Source is missing, failed, or in the wrong Catalog, fix that before asking Tavina to rely on it.

What happens when you upload a document

For document-like files, Tavina should:

  1. create a Knowledge Library Source record;
  2. put the Source in the selected Catalog, or in Library Inbox when no Catalog is selected;
  3. keep compact Source details in chat instead of stuffing the full extracted document into the prompt;
  4. process the Source into durable extracted content and citation-ready chunks when extraction succeeds;
  5. report failed, partial, OCR-needed, encrypted/blocked, unsupported, or empty extraction states instead of pretending the document was fully read.

File types and processing

The current Knowledge Library foundation covers document-like routing for text, Markdown, PDF, and EPUB workflows, plus safe handling for unsupported or failed extraction cases. Some files may need OCR, may be encrypted, may be too damaged to extract, or may only extract partially.

If Tavina says a Source is still processing, failed, partial, OCR-needed, or blocked, treat the answer as limited until the processing state changes.

Photos and visual analysis

For photo uploads in chat, Tavina can analyze many valid JPEG, PNG, GIF, and WebP images. Most valid phone, camera, and drone photos up to 25 MB can be analyzed after upload. Large valid camera or drone photos may be optimized into a smaller analysis copy before Tavina sends them to the vision model; the original attachment should still remain available in the conversation.

When a photo includes image metadata, Tavina may use a bounded summary such as capture date, camera or device, lens or exposure details, dimensions, orientation, and GPS/location when present to improve the visual answer. For Tavina accounts, available image metadata โ€” including GPS/location โ€” is shared with analysis by default because location can be useful for real-world image questions. Users or account operators should still use the available opt-out or redaction configuration when they do not want location metadata included. Tavina may not receive or trust every detail: metadata can be absent, stripped by another app, inaccurate, or intentionally disabled.

If Tavina says an image is still too large for visual analysis after optimization, the image may still remain attached in the conversation, but Tavina may not be able to describe it until you upload a smaller copy. If Tavina says the image type or dimensions are unsupported, try exporting the photo as a standard JPEG or PNG and uploading that copy instead.

Good questions to ask after upload

After uploading a file, ask:

  • “What Source did this create?”
  • “Which Catalog is it in?”
  • “Is processing complete or partial?”
  • “Can you cite the pages or sections you used?”
  • “Did you use targeted retrieval or review the whole Source?”

Protect sensitive files

Do not upload files unless you are comfortable using them with Tavina. Avoid uploading secrets, passwords, payment card numbers, private keys, one-time codes, or records you do not have permission to use.

Photos can also include sensitive context such as GPS/location metadata. If a photo could expose a private location, use your device/app metadata controls or Tavina account opt-out/redaction settings before upload.

If an upload fails

Try:

  1. checking whether the file is supported;
  2. renaming the file with a simple filename;
  3. trying a text or Markdown export when available;
  4. checking whether the file is encrypted or image-only;
  5. using Upload troubleshooting if the problem continues.

Related pages

Scroll to Top