AI Agent with Multifunction Calling using Google Vertex AI > Introduction to AI Agent with Multifunction Calling using Google Vertex AI recipe
Introduction to AI Agent with Multifunction Calling using Google Vertex AI recipe
The AI Agent with Multifunction Calling using Google Vertex AI recipe is based on REST and SOAP APIs. The recipe offers a custom AI agent framework that leverages Google Vertex AI's function calling to dynamically generate, orchestrate, and run Application Integration processes based on user queries and seamlessly pass outputs between steps.
The recipe allows you to specify the user prompt and enter the function declaration in the JSON format. The recipe uses a guide to query other processes based on a function call response.
The process sends a request to Google Vertex AI with the specified function declaration and user prompt. Google Vertex AI recognizes the query, extracts key variables and values, and returns them in the appropriate format. Based on the response received from the function call, the LLM summarizes and describes the extracted parameters.
The user confirms the accuracy of the function call response to make the call to the Application Integration process or doesn't make a function call again. The process requests the first Application Integration process with parameters based on the function-calling response. Based on the first process response, a new request is made to the LLM. Google Vertex AI recognizes the service from which the response came and remaps the fields to the target service according to the instructions. The process makes the next request to the Application Integration recipe based on the function-calling response. The user then receives a message about the success of the recipe call and a response from the second Application Integration recipe.
The recipe contains two processes to perform function calling. You can increase the number of processes or modify the existing processes by updating the recipe parameters.