Function Calling

Translation in Progress

This page is being translated. Content below is a placeholder.

What is Function Calling

Function Calling lets LLMs request execution of predefined functions.

Declaring Functions

Use JSON Schema format:

const functionDeclaration = {
  name: 'read_file',
  description: 'Read the contents of a file',
  parameters: {
    type: 'object',
    properties: {
      path: {
        type: 'string',
        description: 'Path to the file to read'
      }
    },
    required: ['path']
  }
}

Complete Workflow

// 1. Define tools
const tools = [{
  functionDeclarations: [readFileDeclaration, writeFileDeclaration]
}]

// 2. Create model with tools
const model = genAI.getGenerativeModel({
  model: 'gemini-2.5-flash',
  tools
})

// 3. Start chat
const chat = model.startChat()
const response = await chat.sendMessage('Read package.json')

// 4. Handle tool calls
const calls = response.response.functionCalls()
if (calls?.length) {
  const results = []

  for (const call of calls) {
    // Execute tool
    const result = await executeFunction(call.name, call.args)

    results.push({
      functionResponse: {
        name: call.name,
        response: result
      }
    })
  }

  // 5. Send results back to LLM
  const finalResponse = await chat.sendMessage(results)
  console.log(finalResponse.response.text())
}

Summary

• Function Calling = LLM requests tool execution
• Declare tools using JSON Schema
• Execute tools and return results to LLM

Next

Learn about tool interface design: Tool Interface Design →