RAG Retrieval

Overview

The RAG Retrieval node queries a configured vector store with an input query string and returns the top-k most semantically similar document chunks. Use it to inject relevant context into AI model prompts before a battle round executes. Retrieval requires a pre-indexed vector store and valid credentials; if the store is unreachable or the query produces no results above the similarity threshold, the node routes to its error output. Returned documents include content text and metadata (source, score) suitable for direct prompt assembly.

Configuration

Field	Type	Required	Description
vectorStoreId	string	Yes	Identifier of the target vector store. Corresponds to a store registered in the project's integration settings.
topK	number	Yes	Maximum number of document chunks to retrieve. Higher values increase context richness but raise token cost.
similarityThreshold	number	No	Minimum cosine similarity score (0–1) a chunk must meet to be included. Chunks below this score are silently dropped. Defaults to 0.7.
namespace	string	No	Partition key within the vector store. Use to scope retrieval to a specific dataset or tenant when the store is shared.
includeMetadata	boolean	No	When true, each retrieved chunk includes its source metadata object alongside content text. Defaults to false.
embeddingModel	enum	No	Embedding model used to encode the query. Must match the model used at index time. Options: text-embedding-3-small, text-embedding-3-large, text-embedding-ada-002.

Inputs

Port	Type	Description
query	string	The search query string to embed and retrieve against. Typically the current battle prompt or a distilled sub-question.
filter	object	Optional metadata filter object applied server-side before similarity ranking. Shape is store-specific (e.g. {"category": "science"}).

Outputs

Port	Type	Description
documents	array	Ordered array of retrieved chunks. Each item has {content: string, score: number, metadata?: object}. Empty array if no chunks exceed the threshold.
error	object	Emitted when the vector store is unreachable, credentials are invalid, or the query embedding fails. Contains {code: string, message: string}.

Example

{
  "nodeType": "rag_retrieval",
  "config": {
    "vectorStoreId": "vs_lf_science_kb_01",
    "topK": 5,
    "similarityThreshold": 0.75,
    "includeMetadata": true
  }
}