RAG¶

The RAG (Retrieval-Augmented Generation) worker searches a content source for passages relevant to a query, then passes those passages to a language model to produce a grounded response. Use it when answers must be anchored to your own documents or data rather than the model's training data alone.

Parameters¶

Input¶

Field	Description
Query	The question or search term used to retrieve relevant content. Accepts Template Text or Upstream Data
Input Type	The data type of the content to be indexed. Accepts Fixed Value. Defaults to String
Content	The content to search against, supplied by a previous worker. Accepts Upstream Data

Output¶

Field	Description
Output Type	The data type the worker returns. Accepts Fixed Value. Defaults to String
JSON Schema	Optional schema for structured output. Define field types using the Form or JSON editor. Only applies when Output Type is not String

Advanced¶

Field	Description
Splitter	How the content is split into chunks before indexing. Accepts Fixed Value. Defaults to Markdown
Chunk Size	Maximum number of tokens per chunk. Accepts Fixed Number. Defaults to 512
Chunk Overlap	Number of tokens shared between adjacent chunks. Accepts Fixed Number. Defaults to 50
Embedding Model	The model used to generate vector embeddings for retrieval. Accepts Fixed Value. Defaults to Embed Large V3
Search Type	The retrieval strategy. Accepts Fixed Value. Defaults to Similarity
Number Of Chunks	How many retrieved chunks are passed to the model as context. Accepts Fixed Number. Defaults to 10
Results Evaluation	Whether to include a relevance evaluation of the retrieved results. Accepts Fixed Value. Defaults to No

Result¶

Once the worker finishes successfully, it returns the model's response grounded in the retrieved content.

Result (string)