Retriever Node

Performs highly-targeted deterministic information retrieval from configured Document Stores.

Functionality

This node queries one or more configured Document Stores and returns the most relevant chunks using the embedding and vector index configured for those stores. It is a focused path when you only need retrieval (no tool use or multi-step reasoning), which usually uses fewer tokens than routing the same task through an Agent node.

Configuration Parameters

• Document Stores: Specify which pre-configured and pre-embedded Document Store clusters this node queries during execution.
• Retriever Query: Explicitly define the vector semantic text query used to probe the stores. Inject variables (e.g. {{ userInput }}) natively simply using curly braces.
• Output Format:
  • Text: Concatenates the extracted document chunks cleanly into a single long multi-line string.
  • Text with Metadata: Annotates the concatenated chunks with origin references (e.g. Source: employee_handbook.pdf (Page 3)).
• Update Flow State: Persists the retrieved context securely into $flow.state, allowing subsequent arbitrary nodes to reference it effortlessly.

Inputs & Outputs

• Inputs: Expects the vector query string parameter (often sourced dynamically).
• Outputs: Streams exactly the semantic documents formatted properly per the configured Output type.