Split Text
On this page
Purpose
Split Text breaks Message, Data, or DataFrame content into smaller chunks and returns a DataFrame while preserving metadata.
When to use it
- Prepare long text for a retrieval pipeline.
- Break a document into manageable pieces before processing.
- Split structured content with a known text field.
How to use it
- Connect content to Input.
- Set Chunk Size and Chunk Overlap.
- Choose a separator.
- Send Chunks to a DataFrame-aware downstream component.
Configuration
| Setting | What it controls | Recommended starting point |
|---|---|---|
| Chunk Size | Sets the target chunk length. The default is 1000 characters. | Start with 1000. |
| Chunk Overlap | Repeats trailing content in the next chunk. The default is 200 characters. | Start with 200. |
| Separator | Sets the boundary used to split text. The default is a newline. | Use \n\n for paragraph boundaries. |
| Text Key | Selects the text field for Data and DataFrame input. The default is text. | Keep text when the input uses that field. |
| Keep Separator | Retains separators in chunks. It is disabled by default. | Leave it off unless downstream text needs separators. |
Expected result
The component returns a DataFrame of chunks. Missing input and an empty DataFrame return errors. Valid empty text succeeds with a zero-row DataFrame. An individual segment that is larger than Chunk Size can remain larger than the target.
Reference details
Split Text does not embed or index content. Tune size and overlap against the downstream model or retrieval component.