AI Assistant Architecture

Overview

AI Assistant architecture integrates on‑premises and cloud components to provide secure, intelligent responses to natural‑language input, including search queries and access requests.

 

Architecture

The diagram illustrates the AI Assistant architecture. Licenses are validated and users are authenticated on-premises, while AI processing occurs in the Tufin Cloud.

On-premises components

AI Assistant Service

The AI Assistant Service is a local on‑premises service that verifies license status to ensure users are authorized to use AI capabilities, and manages user authentication.

TOS Web Client

The TOS Web Client is the browser user interface integrated with the Tufin Orchestration Suite (TOS) where users submit natural-language requests for AI processing. The TOS Web Client handles interactions between the user and the cloud to ensure protected and compliant data exchange.

Cloud components

TOS Cloud AI Service

The TOS Cloud AI Service is the central cloud-based component responsible for processing natural-language requests. It functions as a secure, isolated service that does not require direct communication with the on-premises environment and communicates through the TOS Web Client.

LLM

The Gemini Flash large language model (LLM), powers natural language understanding and conversion to the appropriate format. For example, it converts search queries to Tufin Query Language (TQL).

Data Storage service

The Data Storage service securely stores natural-language requests, generated responses, and feedback submitted by the user. Sensitive data fields are automatically masked and encrypted to maintain customer data privacy.

How does AI Assistant work?

The diagram below illustrates the high-level data flow of the AI Assistant process when the user submits a natural-language request. Each numbered arrow corresponds to a step in the sequence described in the list that follows.

  1. The user submits a natural-language request in the TOS Web Client.

  2. The TOS Web Client securely sends the natural-language request to the TOS Cloud AI Service, which forwards it to the LLM for processing.

  3. The LLM (Gemini Flash):

    • Interprets the natural-language request

    • Generates the equivalent response in the required format

    • Forwards the generated response to the TOS Cloud AI Service

  4. The TOS Web Client:

    • Pulls the results from the TOS Cloud AI Service

    • Displays the natural-language request, the generated response, and the results, if any, to the user

  5. The user submits feedback, and the TOS Web Client forwards it to the TOS Cloud AI Service.

  6. The TOS Cloud AI Service masks sensitive values, and saves the natural-language request, the generated response, and feedback, in the Data Storage Service.