Document Actions

Click the three-dot menu (⋮) on any document row to access:

Action	Description
Open Document	Navigate to the original source document
View Stats	See detailed document statistics
Re-Sync	Force a fresh sync of this document
Tags	Manage tags assigned to this document
Delete	Remove the document and its vector chunks

Understanding Document Re-Sync

The Re-Sync action reprocesses a document from scratch. When you resync a document, the system:

1. Deletes existing chunks - Removes all vector embeddings for that document from your vector store
2. Re-fetches content - For cloud sources (Confluence, Drive), fetches the latest content from the source
3. Re-chunks the content - Splits the document using your current chunking settings
4. Re-embeds all chunks - Generates new vector embeddings using your current embedding model
5. Stores new vectors - Saves the fresh embeddings to your vector store

When to Re-Sync Documents

Scenario	Why Resync Helps
Changed embedding model	Switched from text-embedding-ada-002 to text-embedding-3-small - old embeddings are incompatible with new ones
Changed embedding provider	Moved from OpenAI to Cohere - embeddings have different dimensions and can't be mixed
Updated chunking settings	Changed chunk size from 512 to 1024 tokens - existing chunks don't reflect new settings
Source document changed	Someone updated the Confluence page or Google Doc - your vectors contain stale content
Previous sync failed	Document shows "Error" status - resync after fixing your API key or config
Changed vector store	Switched from Supabase to Pinecone - need to populate the new store

Important

If you change your embedding model or provider, you should resync all documents to ensure consistent search results. Mixing embeddings from different models produces poor search quality because each model encodes meaning differently.

Enhanced Re-Sync with Contextual Retrieval

When you click Re-Sync on a document, a dialog appears offering two sync modes:

Standard Re-Sync

The default mode that uses your current chunking settings to split and embed the document. This is fast and cost-effective for most use cases.

Enhanced with Contextual Retrieval

This advanced mode uses an LLM to generate a brief context statement (1-2 sentences) for each chunk before embedding. The context situates the chunk within the overall document, dramatically improving retrieval accuracy.

How it works:

1. Document is chunked using your current settings
2. For each chunk, the LLM generates a context prefix like:

   "This chunk is from the Vector Data Loader documentation, specifically the 'OAuth Setup' section under 'Authentication', discussing token refresh mechanisms."

3. The context is prepended to each chunk before embedding
4. When you search, the enhanced embeddings better match relevant queries

Benefits

• ~49% fewer retrieval failures (based on Anthropic's research)
• Better semantic matching for out-of-context queries
• Improved relevance when document structure matters

Cost Estimate

Before processing, the dialog shows:

• Document size in tokens
• Number of chunks to process
• LLM provider and model being used
• Estimated cost (typically $0.01-0.10 for most documents)

Requirements

• LLM must be configured in Settings > LLM Model
• Uses your LLM API credits (not Vector Data Loader quota)
• Works with any configured LLM provider (OpenAI, Anthropic)

When to Use Enhanced Mode

Scenario	Recommendation
Important documents	Use enhanced for key reference docs you search frequently
Long technical docs	Enhanced helps with multi-section documents
General content	Standard mode is usually sufficient
Many documents	Standard for bulk; enhanced selectively for important ones

tip

Contextual retrieval is most beneficial for documents where chunks might be searched out of context. For short, focused documents, standard sync is often sufficient.