LLM API Interaction

Translation in Progress

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

Overview

Before building an Agent, you need to understand how to communicate with LLM APIs.

This chapter uses the Gemini API as an example.

Basic Setup

Installation

npm install @google/generative-ai

Initialization

import { GoogleGenerativeAI } from '@google/generative-ai'

const genAI = new GoogleGenerativeAI(process.env.GEMINI_API_KEY!)
const model = genAI.getGenerativeModel({ model: 'gemini-2.5-flash' })

Conversation Modes

Single-turn

const result = await model.generateContent('Hello!')
console.log(result.response.text())

Multi-turn Chat

const chat = model.startChat()

const response1 = await chat.sendMessage('My name is Alice')
console.log(response1.response.text())

const response2 = await chat.sendMessage('What is my name?')
console.log(response2.response.text()) // "Your name is Alice"

Configuring Tools

This is key for Agents - telling the LLM what tools are available:

const model = genAI.getGenerativeModel({
  model: 'gemini-2.5-flash',
  tools: [{
    functionDeclarations: [
      {
        name: 'read_file',
        description: 'Read file contents',
        parameters: {
          type: 'object',
          properties: {
            path: { type: 'string', description: 'File path' }
          },
          required: ['path']
        }
      }
    ]
  }]
})

Handling Tool Calls

const response = await chat.sendMessage('Read package.json')

// Check if LLM wants to call a tool
const functionCalls = response.response.functionCalls()

if (functionCalls?.length) {
  for (const call of functionCalls) {
    console.log(`Tool: ${call.name}`)
    console.log(`Args: ${JSON.stringify(call.args)}`)
  }
}

Summary

• Use @google/generative-ai SDK
• startChat() enables multi-turn conversation
• Declare tools via tools config
• functionCalls() gets tool call requests

Next

Learn about streaming and System Prompts: Prompt & Streaming →