Tool calling

Tool calling lets a model produce structured arguments for a function your application controls. Your code remains responsible for validating the arguments, running the function, and sending its result back to the model.

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.EUROUTER_API_KEY,
  baseURL: "https://api.eurouter.ai/api/v1",
});

const messages = [{ role: "user", content: "What is the weather in Amsterdam?" }];

const first = await client.chat.completions.create({
  model: "mistral-large-3",
  messages,
  tools: [
    {
      type: "function",
      function: {
        name: "get_weather",
        description: "Get the current weather for a city",
        parameters: {
          type: "object",
          properties: {
            city: { type: "string" },
          },
          required: ["city"],
          additionalProperties: false,
        },
      },
    },
  ],
});

const assistantMessage = first.choices[0].message;
messages.push(assistantMessage);

for (const toolCall of assistantMessage.tool_calls ?? []) {
  const args = JSON.parse(toolCall.function.arguments);

  // Validate args before calling your own trusted function.
  const result = await getWeather(args.city);

  messages.push({
    role: "tool",
    tool_call_id: toolCall.id,
    content: JSON.stringify(result),
  });
}

const final = await client.chat.completions.create({
  model: "mistral-large-3",
  messages,
});

console.log(final.choices[0].message.content);

Safety checklist

Treat tool arguments as untrusted input and validate them against a schema.
Use an allowlist of tools; never execute a function name supplied directly by the model.
Require explicit user confirmation before destructive or costly actions.
Apply authorization checks in the tool implementation, not in the prompt.
Set timeouts and record tool outcomes for debugging.

Tool support varies by model. Check the model catalog before relying on it in production.

Tool calling

Safety checklist

On this page