tool-use

Safely orchestrate external API calls by defining explicit tools with input schemas and using an execution loop that runs handlers, applies risk classification, and returns structured errors to prevent silent failures.

Handle errors during LLM function calling by returning structured errors from the execution loop, allowing the model to retry the tool call or trigger a fallback response without silent failures.

You can use LLM tool use to fetch live data from databases and files by defining deterministic functions as tools, allowing the model to delegate retrieval tasks rather than generating data itself.

Prevent destructive actions when an LLM triggers side effects by applying automatic risk classification to tool calls and prompting for user confirmation before executing high-risk operations.

You need explicit tool definitions for LLM API integration to provide clear selection criteria, specifying the tool name, description, and input schema so the model can accurately delegate work to deterministic functions.

An LLM execution loop fails to return API results if tool handlers do not pass structured outputs back to the model, a step required to inform subsequent text generation or further tool calls.