AI Function Definition and Management Guide

AI Functions and Configuration Guide

AI functions (Function Calling) are actual tools your assistant can use to pull information from databases or integrate with external systems, rather than just chatting.

Working Principle (Function Calling)

When the AI model receives a user message, it checks the descriptions of the functions assigned to it. If the user asks for dynamic information like "What time is tomorrow's sales meeting?", the model indicates that it will call the get_calendar_events function instead of directly generating a reply. Our system executes this function, passes the result back to the model, and the model presents the final response to the user.

Parameter Definition and Validation

When defining functions, you specify what information the AI should send as parameters (using JSON schema). For example, a cargo tracking function might require a tracking_number parameter. The AI extracts this parameter from the message, validates its format, and passes it to the API.

AI Settings: Temperature and Token Limits

You can configure the AI settings for each assistant:

• Temperature: Takes a value between 0.0 and 1.0. 0.0 ensures the assistant provides precise, consistent, and reproducible responses (recommended for operational tasks). 1.0 generates more creative and varied responses.
• Maximum Token Limit: Sets the maximum word/character limit the AI model can generate in a single request, preventing high credit consumption.

Assistant Authorizations

Not every assistant can use every AI function you define. For security, you must select which functions (e.g., department routing, draft writing) are enabled for each assistant individually in the company panel.