Page-level chunking in datastore configuration options
Today, Contextual AI announces the addition of page-level chunking mode to our datastore configuration options, optimizing document processing for page-boundary sensitive content.
Clients can now use the new page level chunking mode, which creates chunks that respect page boundaries unless the maximum chunk size limit is reached. This mode is particularly effective for slide decks, where maintaining page-specific context enhances retrieval accuracy.
The page-level mode joins our existing chunking options (heading-depth, heading-greedy, and simple-length) to provide more granular control over how your documents are segmented for RAG processing.
To get started, simply set chunking mode to page level in your datastore configuration when using our ingest document API or datastore UI.