Manage Retrievers

Create and customize retrievers for Retrieval Augmented Generation (RAG) in Data 360. To see the list of retrievers, navigate to the Retrievers tab in AI Models (formerly Einstein Studio). To learn about a retriever, select View Details from the dropdown menu. Each time you edit a custom retriever and save the changes, a new version is created. You can create multiple versions, but you can have only one version active at a time.

Retrievers adhere to the data governance policies set up for your data. Access checks on the results make sure that the user can securely access the standard object and records. If the AI agent or user calling the retriever doesn’t have access to the source record, it doesn’t appear in the retriever results. For more information see, Best Practices for Agent User Permissions.

• Create an Individual Retriever
  To improve the relevancy of the content returned to a prompt for a particular use case, create a custom individual retriever in AI Models (formerly Einstein Studio). To customize the retriever, select its search index, define filters, and specify what information it returns.
• Create an Ensemble Retriever
  Use ensemble retrievers to get the most relevant information from multiple searches or data sources. In AI Models (formerly Einstein Studio), an ensemble retriever is a collection of individual retrievers. When you run an ensemble retriever, it runs all the individual retrievers, combines their results into a single list, reranks the list according to relevance to the search request, and returns just the most relevant information to the prompt or agent.
• Activate or Deactivate a Retriever Version
  Activate a retriever version in AI Models (formerly Einstein Studio) and use it in prompt templates. Only one retriever version can be active at a time. Activating a new version deactivates the current one. Deactivate a version only if no prompt templates depend on it.
• Edit or Delete a Retriever Version
  Edit retriever versions in AI Models (formerly Einstein Studio) to create versions for testing different configurations and responses within a prompt. You can edit only custom individual retrievers. If no prompt templates depend on a retriever, you can delete it.