Connected AI Agents for Permission-Aware Messaging Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence (AI) models, such as Large Language Models (LLMs), are limited in responding to questions involving personal and enterprise content items like documents, emails, and text conversations due to insufficient training data, and users face difficulties in locating relevant information within messaging applications, leading to inefficient and distracting search processes.
Innovation Solution
A connected AI agent system that integrates a connector service and AI agent across multiple servers, vectorizes user content items, and executes agent objects to provide intelligent responses within messaging environments, leveraging user permissions and management policies to access and generate responses from vector databases.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If users perform traditional text searching in folders or message histories, then they can locate relevant personal and enterprise items, but the process is time-consuming and requires leaving the messaging environment
Solution Approach 1:
The system pre-processes and vectorizes content items (documents, emails, texts, images, videos, audio recordings) into embedded representations and stores them in vector databases before they are needed for searching. This preliminary action enables instant semantic similarity comparisons when users query the AI agent, eliminating the need for time-consuming traditional text searches while maintaining high accuracy in locating relevant information.
Solution Approach 2:
The patent introduces an AI agent as an intermediary between users and their content repositories. The AI agent receives natural language queries from users, performs semantic searches across multiple vector databases containing different types of content, and returns relevant results. This intermediary layer abstracts the complexity of searching through folders and message histories, providing accurate results without requiring users to leave their messaging environment.
2Adaptability or versatility
If public LLMs are used to answer questions, then they provide general knowledge responses, but they cannot access personal and enterprise content items
Solution Approach 1:
The system implements a nested architecture where the AI agent (outer layer) encapsulates multiple specialized components including vector databases, embedding models, and content repositories (inner layers). The AI agent receives user queries and nests subsequent processing steps within this structure, sequentially accessing relevant vector databases and retrieving appropriate content items. This nested design enables the system to maintain the general knowledge capabilities of public LLMs while integrating access to private enterprise content through layered data structures.
Solution Approach 2:
The AI agent is designed as a universal system that can handle multiple types of content (documents, emails, texts, images, videos, audio recordings) through a single interface. It performs multiple functions including semantic search, content retrieval, and response generation across diverse data types. The system uses embedding models to convert different content types into comparable vector representations, enabling the AI agent to universally access and process various personal and enterprise content items regardless of their original format or storage location.
3Adaptability or versatility
If users open multiple applications to perform searches, then they can access different content types, but they lose focus and compete for user attention
Solution Approach 1:
The patent merges multiple content repositories (documents, emails, texts, images, videos, audio recordings) into a unified vector database architecture. Instead of requiring users to open separate applications for each content type, the system combines all these content sources into integrated vector databases that can be searched through a single AI agent interface. This merging maintains comprehensive content access capabilities while eliminating the need for users to switch between multiple applications, thereby preserving their focus and attention.
4Reliability
If traditional LLM training data sources are used, then models are trained on public data, but they lack personal and enterprise content for specialized responses
Solution Approach 1:
The system creates vectorized copies (embeddings) of personal and enterprise content items and stores them in vector databases. Instead of requiring the LLM to be retrained on all private data (which would be complex and resource-intensive), the system copies the essential semantic features of the content into vector representations. These copied vector forms can be efficiently queried and retrieved by the AI agent, providing reliable specialized responses without the complexity of full model retraining on proprietary data.
Data Source
AI summary
Systems and methods are described for a messaging agent for managed multidimensional search based on an electronic message and management policies. A messaging application at a client device can send a new electronic message, such as an email, to a connector service. An attachment can be ingested and stored in a vector database. Then one or more artificial intelligence (“AI”) agents can be selected for responding to the body of the email, such as a query in the body. The responses can be formatted and sent to multiple users, such as a sending user of the electronic message and a recipient user that was copied or also sent the new electronic message. The AI agents can use different AI models, prompts, and vector databases depending on user permissions. This allows for building up vector databases with relevant content items and answering user questions based on those vector databases.