AI Agent Tools

The 4-Step Handshake

How a chat becomes action

1. Tool declaration: Define the tool name and parameters.
2. Reasoning: Chatbot decides to use the tool.
3. The Call: Chatbot returns parameters as a JSON object (not text).
4. Execution: Your app runs a function with the parameters.

(Press Down ↓ to see the code contract)

Step 1: Tool Declaration

{
  "name": "generateMovieScript",
  "description": "Updates the scene editor with a new script",
  "parameters": {
    "type": "object",
    "properties": {
      "scenes": {
        "type": "array",
        "items": { 
          "type": "object", 
          "properties": { "description": "string", "narration": "string" } 
        }
      }
    },
    "required": ["scenes"]
  }
}
                    

Step 2: Chatbot Response

The "Invisible Handshake"

{
  "functionCall": {
    "name": "generateMovieScript",
    "args": {
      "scenes": [
        { "description": "Neon city...", "narration": "Welcome..." }
      ]
    }
  }
}
                    

Step 3: "Catch & Execute" Function

How the app handles the chatbot's request.

async function handleChat(userInput) {
  const result = await chat.sendMessage(userInput);
  
  // 1. Check if Gemini requested a Tool
  const call = result.response.functionCalls()?.[0];

  if (call?.name === "generateMovieScript") {
    
    // 2. EXECUTION: Run the local React function
    handleAssistantScript(call.args.scenes); 

    return "I've updated the editor for you!";
  }
  
  return result.response.text();
}