Section 3 · Data 360 Fundamentals · 20% of exam
3.5 Retrievers
Exam objective
Explain how retrievers connect a search index to a prompt or agent, and identify when a standard or custom retriever is appropriate.
The retriever’s role
A retriever is the component that searches configured data and returns relevant information for a particular use case. In a RAG workflow, it links the prompt or agent experience to the search index. The prompt supplies a query, the retriever searches the index, and relevant results are added to the prompt as context. The language model then generates an answer based on the instructions, user request, and returned material. The retriever does not itself write the final answer, and the model does not automatically inspect every record in Data 360. This distinction is important in scenario questions: the index stores and organizes searchable content; the retriever controls how relevant content is found and supplied.
Default library retrievers
When a supported Agentforce Data Library is set up, Salesforce automatically creates the search index and retriever needed for the library. This quick-start approach is useful when the source is Knowledge or uploaded files and the standard library behavior fits. The library is connected to an agent through an appropriate action, such as Answer Questions with Knowledge. Confirm the library is Ready and make sure the agent is configured to use it. Salesforce provides an option to show sources so that responses can include source information. Standard library setup reduces manual configuration, but students should still understand which source and index the retriever uses. Custom retrievers and advanced use cases
A custom retriever can connect a library to a retriever configured in Data 360. Salesforce Help says that, from a home org, the search index and retriever must first be created and the retriever activated. Only active retrievers appear for selection. Custom retrievers can support more types of Data 360 data and certain companion-org patterns. Salesforce documentation also describes ensemble retrievers, which combine retrieval across multiple sources. Check the current product documentation for availability and limitations: setup details can evolve. A custom retriever is appropriate when the standard library path does not meet the data or retrieval design, not simply because customization is available.
Choose scope and filters deliberately
Retrievers can be designed to focus on a relevant subset of information. This can improve relevance and avoid returning unrelated material. A retriever’s configuration should match the use case, the search index, and the data that the agent is allowed to use. Review filters, object and field inclusion, source freshness, and the user or agent security context. Salesforce documents limitations for some filter behaviors when using the Answer Questions with Knowledge prompt template, so verify whether filters are static or dynamic in the selected setup. If a design needs runtime filtering by a user-selected category, determine whether the supported library configuration meets the requirement or whether a different implementation is needed.
Troubleshoot the retrieval link
If an answer lacks evidence, inspect the retriever’s target index, active status, query behavior, filters, and returned results. A correct index connected to the wrong retriever can still yield poor answers. A retriever may also find semantically related content that does not contain the fact the user needs. In that case, revise the source collection, metadata, query formulation, or retrieval configuration, then retest. Use a set of questions that includes synonyms, exact identifiers, ambiguous terms, and unanswerable requests. When citations are enabled, compare the cited source with the final claim. This helps determine whether the problem is retrieval or generation.
If a scenario asks what connects a prompt to a search index and returns relevant passages, the key component is the retriever.
A custom retriever must be active and available in the right org context. Confirm current support, filtering limitations, and companion-org requirements in Salesforce Help.
Extra concept: Think of the index as the searchable collection and the retriever as the configured search path that selects useful context for a query.
Official references:
- Enhance Agents with Business Knowledge Using RAG https://trailhead.salesforce.com/content/learn/modules/retrieval-augmented-generation-quick-look/augment-prompts-with-relevant-knowledge
- Use a Custom Retriever https://help.salesforce.com/s/articleView?id=ai.data_library_custom_retriever.htm&language=en_US&type=5
- Use the Answer Questions with Knowledge Action https://help.salesforce.com/s/articleView?id=ai.agent_setup_data_sources.htm&language=en_US&type=5
Key takeaways
- A retriever searches an index and returns relevant context for a prompt or agent.
- A standard data library automates its index and retriever; use custom retrievers for supported advanced needs.
- Check retriever target, active status, filters, and actual results when diagnosing weak grounding.
