Name the tool itself, and all associated properties to accurately reflect what their purposes are, as accurately as possible, and write property descriptions to help it know how to populate them. Then in prompt, say something like " - Always and only use the 'addrelative' tool for adding a relative to the caller's account - Always and only use the 'registeruser' tool for trying to register a caller as a user" etc. Also, which LLM model are you using, because they're not all equal when it comes to reliably calling tools. A tool called "checknumber" is pretty vague without providing ample context to the assistant about what it's for. The tool description, property description(s), and prompt should all aim to help the model comprehend what the the tool is for, what the properties should be populated with, and when to call the tool.