Document Index

Indexes the document into an ephemeral vector collection (deleted at the end of the job) so an agent or supervisor can retrieve related passages, even far apart in the text, in addition to the text it normally receives. Unlike a knowledge base, this index only covers the document currently being analyzed, not external reference documents.

Ports

• Input (left, content type).
• Output (right, document index type) — connect it to the dedicated "index" input of an AI Agent or a Supervisor. This port only accepts a single connected source.

Configuration

• Number of passages retrieved (top_k) — 1 to 20, default 4.
• Indexing segment size (chunk_size) — 200 to 4000 characters, in steps of 100, default 900 (distinct from the size used by the Chunking node).
• Embedding model — among the configured embedding models, or "Auto" (default provider) if left unset.

note

If the downstream agent or supervisor is attached to a knowledge base, the knowledge base takes precedence over this index — the two mechanisms serve different needs and can coexist.