Tool calling

Add tool definitions to the tools array, validate model-generated arguments, and execute the function in your application. The two OpenAI-style protocols use different tool shapes:

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_HILINKUP_API_KEY",
    base_url="https://api.hilinkup.com/v1",
)

Protocol	Function definition	Returned call	Tool result
Responses	`type`, `name`, `description`, and `parameters` are direct tool fields	`function_call` item in `response.output`	`function_call_output` with the same `call_id`
Chat Completions	Definition is nested under `function`	`choices[0].message.tool_calls`	`tool` message with `tool_call_id`

Do not change only the endpoint and keep the other protocol’s tool shape.

Treat tool arguments as untrusted input. Enforce authorization and schema validation before execution.

Tool-call support varies by model. Confirm capability with the selected model before production use.

Last modified on August 12, 2026

Introduction

Core capabilities

Reliability