Interface EntityDocumentChunkingConfig

Whether to enable chunking for long-form text content. When enabled, records with text exceeding maxChunkTokens are split into multiple vectors with shared metadata. Default: false (entire rendered template = one vector).

Maximum tokens per chunk. The embedding model's context window is the hard ceiling. Default: 512.

Number of tokens to overlap between consecutive chunks. Ensures context continuity at chunk boundaries. Default: 50.

Chunking strategy:

• "token": split on token boundaries (most precise)
• "sentence": split on sentence boundaries (more readable)
• "paragraph": split on paragraph/double-newline boundaries