Custom retrievers in a Data Library
Key Takeaway
A custom retriever data library uses a retriever that you already created and activated in Data 360. Only active retrievers that exist in the home org appear in the list you choose from.
A custom retriever is one of the data source types for an Agentforce Data Library. You don't have the library build its own search index and retriever. Instead, you point the library at a retriever that already exists in Data 360. Custom retrievers connect agents to more types of data sources in Data 360, including ensemble retrievers. They also connect companion orgs to data libraries and search indexes configured in the home org.
Data sources for a data library
A data library can use Knowledge articles, uploaded files, or a custom retriever as its data source. With Knowledge articles or uploaded files, the data is indexed in Data 360, where a search index and a retriever are created for you automatically. With a custom retriever, you bring a retriever you already built.
When you connect a data library to an AI agent, the Answer Questions with Knowledge action uses the library's search index and retriever to find answers.
A data library can't hold multiple data sources at once. You also can't change a library's data source after you choose it, so to use a different source you create another data library.
Data 360 must be set up to use Agentforce Data Libraries. Data libraries consume Data 360 credits.
New web search data libraries aren't supported. Existing web search data libraries are read-only, which means you can still view and use them.