Chunking

Splits the document into segments before sending it to the AI models downstream — needed for documents too long to fit into a single call. This node's settings apply to everything downstream in the graph (they override the pipeline's default values).

Ports

• Input and Output (left/right), content type. Connections leaving this node are drawn as dashed lines in the editor, to visually flag that chunking is configured upstream.

Configuration

• Segment size (chunk_chars) — 0 to 50,000 characters, in steps of 500, default 12,000. A value of 0 disables chunking (the whole document is sent in a single call).
• Overlap (chunk_overlap) — 0 to 5, default 1: number of overlapping blocks kept between two consecutive segments, to avoid splitting information across a segment boundary.