Federated RAG Access Control Across Distributed Vector Databases

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing retrieval-augmented generation systems face challenges in managing access control for large and distributed electronic files, as they often exceed the capacity of individual machine learning models and lack efficient methods to handle federated retrieval across multiple locations.

Innovation Solution

A system and method for federated retrieval-augmented generation (FRAG) using vector databases, which converts content into vector embeddings, performs access control, and generates replies based on user queries, ensuring only authorized users can access relevant information through a centralized agent controlling individual sub-systems.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If retrieval-augmented generation is used for large and distributed electronic files, then the system can handle files too numerous and large for individual machine learning models, but access control management becomes complex and difficult to implement across federated locations

Engineering Contradiction:
Improvecapability to handle large and distributed electronic filesVSAvoidaccess control management complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system segments the federated retrieval-augmented generation system into multiple sub-systems, each responsible for specific vector databases and electronic files at different locations. Each sub-system independently manages its own access control for local files, while a centralized agent coordinates across sub-systems. This segmentation reduces access control complexity by localizing management responsibilities to individual sub-systems rather than requiring centralized management of all files across the federation.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

A centralized agent acts as an intermediary between users and the federated sub-systems. The agent receives user queries, determines which sub-systems have relevant files, and coordinates access requests across the federation. This intermediary simplifies access control by providing a single point of coordination that manages cross-location access without requiring each sub-system to directly communicate access control policies with every other sub-system.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Area of stationary object

If federated retrieval-augmented generation is implemented across multiple locations, then the system can access distributed electronic files, but the number of access control verification points increases

Engineering Contradiction:
Improvecoverage of distributed electronic filesVSAvoidnumber of access control verification points
Core Design Contradiction:
Area of stationary objectVSDevice complexity

Solution Approach 1:

The system divides the federated network into autonomous sub-systems, each managing its own vector databases and electronic files. Each sub-system implements access control locally for its own files, reducing the need for centralized verification at every location. The segmentation allows each sub-system to independently verify access for its local files while the centralized agent only needs to coordinate between sub-systems, not within each sub-system.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Each sub-system performs self-service access control verification for its own local files and vector databases. When a user query is routed to a specific sub-system, that sub-system independently verifies the user's access rights to its local files without requiring external verification from other sub-systems or the centralized agent. This self-service approach reduces the total number of verification points by eliminating redundant cross-location verification for local resources.

Inventive Principle:
Principle #25Self-service

3Ease of operation

If individual sub-systems perform retrieval-augmented generation on individual vector databases, then the system can operate autonomously at each location, but coordination and control across the federation becomes challenging

Engineering Contradiction:
Improveautonomous operation at each locationVSAvoidcoordination and control across federation
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The federated system is segmented into autonomous sub-systems that independently perform retrieval-augmented generation on their own vector databases. Each sub-system maintains operational independence, processing queries locally when possible and maintaining its own access control policies. This segmentation preserves ease of operation at each location while the centralized agent provides high-level coordination for queries spanning multiple locations.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

A centralized agent serves as an intermediary that coordinates between autonomous sub-systems without interfering with their local operations. The agent receives user queries, determines which sub-systems should handle them, and routes queries appropriately. This intermediary enables coordination across the federation while preserving sub-system autonomy, as the agent only manages inter-sub-system communication and does not micromanage individual sub-system operations.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentUS12488136B1Systems and methods for access control for federated retrieval-augmented generation
Publication Date: 2025.12.02 INSTABASE INC
  • US12488136B1 patent drawing
  • US12488136B1 patent drawing
  • US12488136B1 patent drawing

AI summary

Systems and methods for providing access control for federated retrieval-augmented generation (FRAG) using a set of vector databases are disclosed. Exemplary implementations may: receive user input representing a user query; verify whether the user has access to particular electronic files; generate agent queries based on tasks included in the user query; select subsets of the vector embeddings based on the agent queries; provide the subsets to one or more machine learning models as context; provide prompts to the one or more machine learning models; obtain replies in response to the prompts; generate a reply to the user query; effectuate a presentation of the reply, and/or perform other steps.