Default retrievers versus custom retrievers
Key Takeaway
Default retrievers are legacy and can't be edited. To ground a prompt template in only part of a search index, build a custom individual retriever in AI Models that filters on a field the index defines for filtering. Then activate it and add it to the template as a resource.
A retriever searches a Data 360 search index. It passes the most relevant results to agents, prompt templates, flows and other features, which use them as grounding. Data 360 used to create a default retriever automatically for every new search index, but it no longer does. You now build retrievers yourself in AI Models (formerly Einstein Studio) and shape each one to a specific use case.
What a retriever does
Search indexes are often built for general use, such as indexing an entire knowledge base, so they hold more than any one task needs. A retriever applies that multi-purpose index to one use case. For example, a retriever can return only one category of articles, only content from the last 90 days, or only content for one geographic region.