#Tool Calling

Tool calling lets a model produce structured function arguments. Your application must validate the arguments, authorize and execute the operation, and return the result. The model never executes your function by itself.

#Choose a protocol

Chat Completions declares functions under tools[].function. Responses places name, description, and parameters directly on a type: "function" tool. Do not copy request bodies between protocols.

#Safe execution checklist

1. Validate arguments against a strict schema.
2. Check the end user's permission.
3. Apply timeouts and resource limits.
4. Require confirmation for writes or destructive actions.
5. Make repeated tool-call IDs idempotent.
6. Return structured failures instead of pretending success.

A successful text prompt does not prove tool calling works. Verify the exact model, protocol, and client combination before production use.

See the Chat Completions guide, Responses guide, and API Reference.