Prepend fields to add context to chunks
Key Takeaway
Prepend fields add the values of up to two fields, such as Title or Product Name, to each chunk. A chunk that doesn't say what it is about on its own then carries that context, so it is easier to match to a user's question.
When Data 360 builds a search index, it splits content into chunks and turns each chunk into a vector that retrievers search for RAG and agent workflows. A chunk on its own may not say what it is about. Prepend fields give each chunk that context by adding the values of fields you select.
How a search index turns content into chunks and vectors
When Data 360 builds a search index, it breaks the source data into semantically related chunks. It then generates one vector for each chunk, and that single vector represents all the content in the chunk.
In RAG and agent workflows, Agentforce agents and prompt templates use retrievers to find relevant chunks by searching these vectors.
A single chunk, such as a list of troubleshooting steps, may not say what it is about on its own.