Back to the catalogue
IngestionRetrieval
Semantic Chunker
Splits documents on structure rather than character count, so a passage retrieved is a passage that makes sense.
Chunking on a fixed character count cuts a clause in half and hands the model two fragments, neither of which says what the clause said.
Ragenta chunks on headings, list boundaries and table edges, with a configurable size and overlap per knowledge base. You can see the chunks a document produced before anyone asks a question against it.