v21

latestOpenAPI 3.1.0raw.githubusercontent.com2025-05-22122299.2 KB

Late Chunker

Implements the late chunking strategy. Encodes the entire text, then splits recursively, deriving chunk embeddings from the full document embedding.

post/v1/chunk/late

Successful Response: A list of LateChunk objects.

textstring

The actual text content of the chunk.

start_indexinteger

The starting character index of the chunk within the original input text.

end_indexinteger

The ending character index (exclusive) of the chunk within the original input text.

token_countinteger

The number of tokens in this specific chunk, according to the tokenizer used.

embeddingnumber[] nullable

Optional embedding vector (list of floats) for the entire chunk, derived from the full document embedding.