Google Research Introduces Agentic RAG Framework for Gemini Enterprise Agent Platform

Google Research team has introduced a new agentic RAG framework, built into the Gemini Enterprise Agent Platform, which powers a feature called Cross-Corpus Retrieval, now in public preview. The target is a known failure mode in enterprise search. Standard single-step RAG was not built for multi-source, multi-hop queries.

Ask “What are the specs of the server used in Project X?” The system may find a document naming a server ID. It will not know to take that ID and search a second database for specs. The answer comes back partial, or as “not found.” Agentic RAG plans, reasons, and iteratively interacts with data sources.

It handles complex queries to increase dependability and accuracy. Google’s version is the Gemini Enterprise Agent Platform-hosted Cross-Corpus Retrieval powered by Agentic RAG. Like other multi-agent RAG frameworks, it uses agents that work together.

Unlike them, it adds a sufficient context check before generating a response. Compared to standard RAG, it increases accuracy on factuality datasets by up to 34%. Google’s research team also tested it on proprietary internal datasets.

It reports better grounding and improved reasoning accuracy on domain-specific tasks. Think of it as an organized research department, not one search engine. A “Vanilla” RAG system just matches your question to documents.

An LLM then generates a response from those matches. The multi-agent framework splits the job into specialized roles. The Orchestrator decides the request is not a one-step job and delegates.

The Planner Agent maps the information pathways across data sources. The Query Rewriter turns a vague request into several precise search queries. The Search Fanout Agent sends those queries to various retrieval sources.

Finally, an LLM aggregates the collected context into a response. The key difference is persistence. The framework knows when it is missing information and keeps searching.

This stops the model from guessing when the first search is empty. It also avoids a premature “I don’t have enough information.” That persistence comes from the Sufficient Context Agent, a new component in Google’s framework. Consider a doctor asking for a patient’s discharge medications, dietary restrictions, and allergic reactions.

In Phase 1 , Orchestration, the Root Agent parses the request and delegates. The Planner Agent targets Pharmacy, Nutrition, and Clinical Notes. The Query Rewriter breaks the long request into simple, searchable questions.

In Phase 2 , Search, the RAG Agent runs all query fanouts at once. It finds medications and diet, but no allergy mention. A Vanilla RAG system might stop here with an incomplete answer.