Google Research Boosts RAG Accuracy with Iterative Agentic Context Search

GoogleGoogle

Google Research and Google Cloud introduced a new agentic RAG framework designed to handle complex enterprise queries. This framework employs a multi-agent workflow that iteratively searches for sufficient context, improving accuracy beyond standard Retrieval-Augmented Generation (RAG). It aims to deliver more dependable responses by preventing the AI from guessing when information is incomplete across multiple data sources.

Google Research and Google Cloud introduced a new agentic RAG (autonomous AI) framework, available as a public preview in the Gemini Enterprise Agent Platform. This multi-agent workflow goes beyond standard Retrieval-Augmented Generation (RAG) by breaking down complex enterprise queries, iteratively searching for sufficient context. A key innovation, the Sufficient Context Agent, evaluates retrieved information and identifies missing pieces.
Accuracy increase (factuality datasets)
Up to 34%
Cross-corpus retrieval accuracy
90.1%
Latency (cross-corpus vs. single-corpus)
Within 3%
Key innovation
Sufficient Context Agent
Availability
Public preview

Standard RAG struggles with multi-source, multi-hop queries, yielding incomplete answers. This new framework plans, reasons, and iteratively interacts with data sources until sufficient context is gathered. Experiments showed it increases accuracy on factuality datasets by up to 34% and achieves 90.1% in cross-corpus retrieval with similar latency.

The new agentic RAG framework is available as a public preview in the Gemini Enterprise Agent Platform. This capability ensures AI-generated responses are auditable, traceable, and grounded, enabling more dependable AI systems.

View the full update on research.google
Google Research
Google Research
@GoogleResearch
X

Introducing our new agentic RAG framework. A collab with Google Cloud, our multi-agent workflow goes beyond standard RAG by breaking down complex enterprise queries & iteratively searching for sufficient context before generating dependable responses. 📜→https://t.co/A8l499bLrj https://t.co/5fZT49j8TL

289retweets2.4klikes
View on X

Still wondering? A few quick answers below.

Google's new agentic RAG framework is a multi-agent system developed by Google Research and Google Cloud. It is designed to handle complex enterprise queries by autonomously planning, reasoning, and iteratively searching for sufficient context across multiple data sources before generating a response.

Unlike standard RAG, which typically performs a single retrieval step, this agentic RAG framework uses multiple specialized agents and an iterative search process. It includes a Sufficient Context Agent that evaluates retrieved information and identifies missing pieces, prompting further searches until enough context is gathered for a dependable answer.

The Sufficient Context Agent acts as a quality-control inspector within the framework. It evaluates retrieved information, an intermediate draft response, and identifies any missing pieces compared to the original query. If context is insufficient, it provides specific feedback to trigger further iterative searches.

Experiments showed the agentic RAG framework increases accuracy on factuality datasets by up to 34% compared to standard RAG. It also achieved 90.1% accuracy in cross-corpus retrieval scenarios, where the system must select from multiple data sources, with latency comparable to single-corpus versions.

The new agentic RAG framework is available as a public preview offering within Google's Gemini Enterprise Agent Platform. This platform provides the environment for building, scaling, and governing AI agents, including this advanced RAG capability.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update