​
Query Node in Giselle

The Query Node is a powerful component in Giselle designed to search and retrieve information from connected data sources. Its primary function is to execute a query (such as a question or a search term) against a dataset and return the most relevant results.

Currently, the Query Node is optimized to work with the GitHub Vector Store Node, making it a crucial building block for creating Retrieval-Augmented Generation (RAG) applications directly within your Giselle workflows.

​
How to Use the Query Node

Setting up a Query Node involves adding it to your canvas, connecting a data source, and running a query.

​
1. Add a Query Node

From the node toolbar at the bottom of the canvas, select the Query Node, which is identifiable by its magnifying glass icon, and place it in your workspace.

​
2. Connect a Data Source

The Query Node requires an input from a data source to function. You must connect the Output of a data source node (like the GitHub Vector Store Node) to the Input of the Query Node. Until a data source is connected, the node will indicate that it is waiting for a connection.

​
3. Write and Run a Query

Once a data source is connected, you can configure the Query Node:

  • Enter a Query: In the configuration panel, type your question or search term into the “Query” input field.
  • Run the Query: Click the Run Query button to execute the search against the connected data source.

​
4. Review the Results

After the query runs, the results will be displayed directly in the node’s panel. The results typically include:

  • A summary of the number of results found.
  • A list of relevant “chunks” of text retrieved from the source documents.
  • A similarity score (e.g., 58%) for each chunk, indicating its relevance to your query.

You can expand each chunk to view its content and metadata.

​
Core Use Case: Retrieval-Augmented Generation (RAG)

The primary purpose of the Query Node is to enable Retrieval-Augmented Generation (RAG) workflows. RAG enhances the capabilities of Large Language Models (LLMs) by providing them with relevant, up-to-date information from your own data sources before they generate a response.

A typical RAG workflow in Giselle looks like this:

  1. Data Source (GitHub Vector Store Node): Ingests and vectorizes your documentation from a GitHub repository, making it searchable.
  2. Query Node: Takes a user’s question (the query) and retrieves the most relevant text chunks from the vector store.
  3. Generator Node: Receives the original question and the retrieved information from the Query Node’s output. It uses this combined context to generate a more accurate, detailed, and factually grounded answer.

​
Output of the Query Node

The Result output of the Query Node contains the collection of retrieved data chunks. This structured data can be passed to subsequent nodes—most commonly a Generator Node—to serve as dynamic context for text generation.

For more information on how to set up a data source for querying, see our documentation on the GitHub Vector Store Node.