AI Assistant Flow Parameter Separation for Secure Tool Calls
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI assistants require significant compute resources, bandwidth, and tokens for complex parameter interactions, leading to inefficiencies and potential data security breaches, while manual setup and coding hinder widespread adoption.
Innovation Solution
A system and method for passing determined parameters in AI flows, using a workflow copilot to separate parameters into LLM-configured and deterministic sets, reducing token usage and enhancing data security by hiding sensitive information from the LLM.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional function calls are used to transfer data back and forth for complex parameters, then AI assistants can access structured data and generate responses, but significant quantities of tokens and computer resources are consumed
Solution Approach 1:
The patent segments parameter passing into two distinct approaches: traditional function calls for simple data retrieval and new deterministic parameter passing for complex parameters. This segmentation allows the system to minimize LLM token usage by using direct parameter passing where applicable, while retaining function calls only when necessary, thereby reducing overall computational resource consumption.
Solution Approach 2:
The patent introduces deterministic parameters as an intermediary mechanism between the AI assistant and the data source. These deterministic parameters act as a mediator that enables complex parameter transfer without requiring full LLM involvement, thus reducing token consumption while maintaining data accessibility.
2Adaptability or versatility
If local data is returned to the AI assistant, then up-to-date and private data can be used, but data security policies and privacy regulations may prevent such data from being transferred
Solution Approach 1:
The patent extracts sensitive local data from the prompt sent to the LLM. By separating and removing sensitive information before LLM processing, the system enables use of local and private data while preventing its transmission to external AI assistants, thus addressing data security concerns and regulatory requirements.
Solution Approach 2:
The patent uses deterministic parameters as an intermediary layer that allows local data to be accessed and used without being transmitted to the LLM. This mediator enables private data utilization while maintaining data security by keeping sensitive information local and unshared with external AI services.
3Adaptability or versatility
If manual setup and coding are required, then AI assistants can be configured with specific parameters, but they cannot be commonly used by individuals without coding ability or time
Solution Approach 1:
The patent implements self-service through automatic parameter determination. The system automatically determines which parameters to pass deterministically based on the task requirements, eliminating the need for users to manually configure complex parameter settings. This automation maintains configuration capability while making the system accessible to users without coding skills.
Solution Approach 2:
The patent changes the approach to parameter configuration from manual coding to automated determination. By using deterministic parameter passing that is automatically selected and configured by the system, the patent maintains full configuration capability while dramatically improving ease of operation for non-technical users.
4Adaptability or versatility
If multiple interactions are used on the back end, then AI assistants can process complex queries, but they become slow and error-prone
Solution Approach 1:
The patent segments the query processing into two paths: simple queries handled by direct function calls and complex queries handled by deterministic parameter passing. This segmentation reduces the number of back-end interactions required for complex parameters, thereby improving response time while maintaining the ability to process complex queries effectively.
Solution Approach 2:
The patent extracts complex parameter handling from the LLM interaction chain and processes them deterministically. By taking out complex parameter processing from sequential LLM calls and handling it through direct deterministic passing, the system reduces processing steps and eliminates errors associated with multiple LLM interactions, thereby improving speed.
Data Source
AI summary
A system may generate a software recipe, which includes triggers and actions where the actions include one or more function tool calls. The system may determine a first set of parameters for the one or more function tool calls and a second set of parameters for the one or more function tool calls, where the first set of parameters is withheld from one or more artificial intelligence (AI) models and the second set of parameters is determined by the one or more AI models. In some implementations, the system may merge the first set of parameters and the second set of parameters for the one or more function tool calls, which use the first set of parameters and the second set of parameters to perform the one or more actions, and the system may execute the software recipe using the one or more function tool calls.