Chunker node

Configure a Chunker node to split partitioned document elements into chunks sized for embedding models, improving retrieval precision in RAG applications.

Learn about the available chunking strategies.

Node	Description
Chunk by Character	Combines elements within hard and soft character limits, without using section or page boundaries.
Chunk by Title	Preserves section boundaries by starting a new chunk at each Title element.
Chunk by Page	Preserves page boundaries, ensuring no chunk spans more than one page.
Chunk by Similarity	Groups topically similar consecutive elements into chunks using an embedding model.
Contextual chunking	Enrichment node placed after a chunker. Uses an LLM to prepend a short context summary to each chunk, preserving chunk count and IDs.
Chunk by Table Merging	Uses an LLM to detect and merge tables that span adjacent pages into a single coherent table element.

Workflow placement

• Always place Chunker nodes after a Partitioner node.

• Chunker nodes are usually placed after any Enrichment nodes present in the workflow.

  An exception is the Named Entity Recognition (NER) enrichment node, which should be placed after any Chunker nodes.

• The chunk_by_table_merging node is often paired with a table2html enrichment earlier in the workflow.

• As the chunk_by_contextual node enriches existing chunks, it should be placed after one of the text chunkers (such as chunk_by_character, chunk_by_page, chunk_by_similarity, or chunk_by_title) in the workflow.