Struct SemanticSplitter 

pub struct SemanticSplitter<E> { /* private fields */ }

Semantic chunker

Cuts a chunk where the similarity between adjacent sentences drops below breakpoint_threshold; forces a break when the accumulated length exceeds max_chunk_size.

Implementations

impl<E: Embeddings> SemanticSplitter<E>

pub fn new( embeddings: E, breakpoint_threshold: f32, max_chunk_size: usize, ) -> Self

Creates a semantic chunker

Arguments

• embeddings - the embedding model
• breakpoint_threshold - the breakpoint threshold for adjacent-sentence similarity (0.0–1.0; lower is harder to break)
• max_chunk_size - the maximum characters per chunk

pub fn with_defaults(embeddings: E) -> Self

Creates with default parameters (threshold=0.5, max=1000)

pub async fn split_text( &self, text: &str, ) -> Result<Vec<String>, EmbeddingError>

Chunks text asynchronously

pub async fn split_document( &self, document: &Document, ) -> Result<Vec<Document>, EmbeddingError>

Chunks a document asynchronously, preserving metadata (writes the chunk index)