Segment¶
Split documents into text chunks of a target length for extraction or retrieval.
For splitting by headings, use Hierarchical Document Splitting. This node controls the length and overlap of text that can be split.
Configure chunking¶
Bind a nonempty upstream documents array.
Setting |
Description |
|---|---|
Chunk size |
Configuration default: 800; target length for splittable text |
Overlap |
Configuration default: 100; characters repeated between adjacent chunks |
Chunk size must be positive. Overlap must be at least 0 and less than chunk size. Length is measured in characters, not model tokens.
Preserve document structure¶
Tables and code blocks retain their structure. Markdown records containing code blocks can also exceed the target length. Images and audio/video segments are handled according to their own structure.
If the next node has an input length limit, check the actual length of large tables and code blocks.
Use the output¶
The output documents contains chunks with source metadata and chunk positions. Connect it to text embedding, extraction, or knowledge base writing nodes.
Example: chunk a product manual¶
Flow: General Document Parsing → Segment → Embed and Add to Knowledge Base.
Set chunk size to 800 and overlap to 100. Check that split sentences retain enough context, and inspect long tables and code blocks separately.