New RAG Features in OCI Generative AI Agents

OCI Generative AI Agents includes the following new features and enhancements:

More detailed citations

In addition to the existing relevant information, the citation feature returns document title and pages, providing more context.

Optional instruction field for agent's responses

For flexibility and control, you can add an optional instruction for an agent's Retrieval-Augmented Generation (RAG) responses. For example, you can ask the agent to answer the prompts in a certain way.

Meta Llama LLM agreement terms

In compliance with legal requirements, when you create an agent, you must agree with the Meta Llama LLM terms.

Optional multi-modal parsing feature

For an agent to include information from charts and graphs of a data source's documents, you can enable the multi-modal parsing feature when you create or edit the data source.

Knowledge base size information

You can get the storage size of a knowledge base through the GetKnowledgeBase API or by navigating to the detail page of that knowledge base in the Console.

Long running ingestion jobs

For large-scale data ingestion jobs that run longer than 24 hours, Generative AI Agents supports resource principals. Use this feature for ingesting large amounts of content.

Citation link override through custom Object Storage metadata

For citation management, you can override Object Storage citation links to custom URLs, by adding metadata to objects. For each object in Object Storage, add a metadata with the name, customized_url_source, and for the value, enter the custom URL.

For information about the service, see the Generative AI Agents documentation.