Skip to main content
Document knowledge bases allow you to store and retrieve information from files, notes, and created text. Documents are indexed using VIVI’s RAGDB system, which breaks content into chunks and uses cosine similarity scoring to return the most relevant results. This makes documents ideal for structured reference material, guides, and knowledge you want your agent to use consistently.

How does it work?

Once you’ve created your Document knowledge base, your index will automatically begin provisioning. You can view the index status in the banner at the top of the page, showing:
  • Number of documents processed
  • Number of documents processing
  • Number of documents that failed ingestion
You can add documents in three ways:

Cloud

Connect directly to your cloud storage provider:
  • Google Drive
  • Dropbox
  • OneDrive Personal
  • OneDrive Work/School
Once connected, browse and import documents. After import:
  • Documents are indexed
  • Status will change from Pending to Processed once complete

Create

Use the built-in editor to write and save new documents directly in VIVI. Perfect for:
  • Short guides
  • FAQs
  • Reference notes
Created documents are indexed like uploaded/cloud-based ones.

Upload

Add files directly from your computer using one of two methods:
  • Via Browser: Best for small files (< 1GB). Drag-and-drop or click to upload.
  • Via Storage Explorer: Best for large files (> 1GB). Generates a temporary token for use with Azure Storage Explorer, enabling:
    • Faster transfer
    • Bulk uploads
    • Third-party collaborator access
Note: After uploading via Storage Explorer, click Sync Files to complete the process.

Best Practices

  • Use Cloud if files are frequently updated — changes can be re-indexed without re-uploading
  • Use Create for text-based additions like SOPs, notes, or FAQs
  • Use Upload for local files or large file sets; prefer Storage Explorer for efficiency
  • Organize content into folders for easier retrieval
  • Verify indexing status — ensure files show as Processed before use in production
  • Remember: Documents are indexed in chunks, so while great for retrieval, they won’t isolate metadata like author names or timestamps
I