One function calling OpenAI, Claude, or Gemini

Every provider has its own client, its own request shape, and its own error format, and I'm tired of writing the same logic three times.

IntermediateRead 6 minDoing: a few hours, most of them spent matching each provider's exact field and key names8 steps

▶ Start

What you'll have at the end

A single function that takes a prompt and a provider name and returns the same reply shape for OpenAI, Claude, and Gemini.

Before you start

You need

A script that already sends a plain prompt to at least one of OpenAI, Claude, or Gemini and gets text back.

Not covered

This builds one call, one tool, one result, and one final answer; it doesn't build the multi-turn loop that keeps calling tools back and forth until the model is done, which is its own recipe once you need it.

Leans on

8 steps

Where to go next

Test your API wrapper without spending real tokens

when you want to test call_with_tool's three branches without spending real tokens on every run

Checked 15 Aug 2026

Part of the LLM APIs and Integration cookbook