RAG Query Node

The RAG Query node performs a vector similarity search against one or more Knowledge Base collections and returns the most relevant document chunks. Unlike the KB Search node, the RAG Query node returns raw search results without generating an LLM response -- giving you full control over how the results are used downstream.

When to Use

• You want to retrieve relevant document chunks and pass them as context to a separate LLM Response node.
• You need fine-grained control over the RAG pipeline (search parameters, result filtering, custom prompting).
• You want to combine results from multiple Knowledge Base collections in a single query.
• You need access to the raw relevance scores and source metadata.

Configuration

Property	Description	Default
Knowledge Base Collections	One or more collections to search. Select from your available KB collections.	Empty (required)
Query Template	The search query. Use {{last_message}} for the user's latest message, or write a custom query with variable interpolation.	{{last_message}}
Output Variable	The session variable to store the search results.	rag_results

Advanced Settings

Property	Description	Default
Top K Results	Number of top results to return.	5
Min Score	Minimum relevance score (0-1). Results below this threshold are excluded.	0.7
Include Source References	Whether to include source document metadata (file name, page number) in results.	false

How It Works

1. The query template is evaluated with current session variables.
2. The query text is converted to a vector embedding.
3. The embedding is compared against all document chunks in the selected collections using cosine similarity on pgvector.
4. The top K results above the minimum score threshold are returned.
5. Results are stored in the output variable as an array of objects, each containing the chunk text, relevance score, and source metadata.

Output Format

The output variable contains an array of result objects:

[
  {
    "content": "The return policy allows returns within 30 days...",
    "score": 0.94,
    "source": "return-policy.pdf",
    "page": 3
  },
  {
    "content": "Refunds are processed within 5-7 business days...",
    "score": 0.87,
    "source": "faq.md",
    "page": null
  }
]

RAG Query vs KB Search

Feature	RAG Query	KB Search
Returns raw chunks	Yes	No (generates LLM answer)
Generates LLM response	No	Yes (built-in)
Multiple collections	Yes	Yes
Custom downstream prompting	Yes (via separate LLM node)	Limited (built-in prompt)
Output	Array of chunks with scores	Generated text answer

imageRAG Query node config panel showing Knowledge Base collection multi-select, query template field with variable interpolation, output variable name, Top K results slider, Min Score threshold, and Include Source References toggle

RAG Query configuration panel

Handles

Handle	Direction	Description
Input	In	Receives execution from the previous node
Output	Out	Continues to the next node after results are retrieved

TIP

For most use cases, the KB Search node is simpler since it handles both retrieval and LLM generation in one step. Use RAG Query when you need more control over the pipeline.