Author(s): Haixi Li

Originally published on Towards AI.

Lesson 1

When you call an LLM API normally, you send messages and get text back. With tool use, you also send a list of tool definitions — name, description, and expected inputs (as a JSON schema). The model can then respond not with plain text, but with a structured request like “call get_weather with {city: 'Tokyo'}." Your code executes that function, and you send the result back to the model so it can continue.

After the intro, the article explains that the model itself doesn’t execute tools—it only decides which tool to call and with what arguments, while your code performs the actions. It lays out the basic tool-call loop (send prompt + tools, model returns either text or a tool-use request, your code runs the function, you send back the tool result, and the model either responds or chains another tool call). It emphasizes why tool descriptions are critical because the model relies on the name and description to choose correctly, then provides a hands-on exercise: define a simple tool like add(a, b), have the model request tool use, compute the result manually, and feed it back so the model can produce the final answer. It extends the exercise to chaining tools (e.g., multiply after add) and walks through a concrete implementation using an API with tool calling, highlighting the “while True” loop as the seed of agentic workflows, concluding with a small exercise to add a subtract tool to verify understanding.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI