Tool calling should be used as a strict boundary between the AI model and application data. Instead of exposing generic or broad access, each tool should target a specific defined use case (e.g., public profile lookup vs. salary access).

Registering multiple small, purpose-specific tools rather than a single large one reduces ambiguity in tool selection and limits the risk of unintended data exposure.

The model is more reliable when each tool has a clear and focused responsibility.

Example:

var $chatHelper:=$client.chat.create("You are an HR assistant.") $chatHelper.registerTool(cs.Tool_GetPublicInfo.new()) $chatHelper.registerTool(cs.Tool_GetPrivateInfo.new())

Separating tools by data sensitivity or business domain improves tool-calling accuracy.