Retriever prefilters for scoping search results
Key Takeaway
A prefilter limits which content a retriever's search can return. You define the condition on the retriever in AI Models, and the prompt template supplies the filter value at run time.
A prefilter narrows a retriever's search to content that meets a condition, such as content for one country or one product, or only active Knowledge articles. The retriever's results become grounding for a prompt template, so a prefilter controls which content can end up in that grounding.
How a retriever grounds a prompt template
You create retrievers in AI Models, which was formerly called Einstein Studio. In Prompt Builder, you then add a retriever to a prompt template as a resource.
A retriever searches knowledge that is indexed in Data 360. It adds the most relevant results to the prompt template, which gives the template relevant, specialized grounding information.