Building AI Agents

Tool Calling

Tools allow AI agents to interact with the real world — search the web, query databases, or call APIs.

What is a Tool?

A tool is a function with:

Name — Unique identifier
Description — What it does (for the LLM)
Parameters — Input schema (JSON Schema format)
Implementation — The actual function

Tool Definition Format

Most AI frameworks use JSON Schema:

{
    "name": "get_weather",
    "description": "Get current weather",
    "parameters": {
        "type": "object",
        "properties": {
            "city": {"type": "string"}
        },
        "required": ["city"]
    }
}

Execution Flow

LLM decides to call a tool
Extracts tool name and arguments
Your code executes the tool
Result sent back to LLM
LLM incorporates result in response

Try It Yourself

Run this code example to practice what you've learned.

example.py

1from typing import Any
2import json
3
4# Define tools with schemas
5TOOLS = [
6    {
7        "name": "get_weather",
8        "description": "Get the current weather for a location",
9        "parameters": {
10            "type": "object",
11            "properties": {
12                "city": {
13                    "type": "string",
14                    "description": "The city name"
15                },
16                "unit": {
17                    "type": "string",
18                    "enum": ["celsius", "fahrenheit"],
19                    "default": "celsius"
20                }
21            },
22            "required": ["city"]
23        }
24    },
25    {
26        "name": "search_web",
27        "description": "Search the web for information",
28        "parameters": {
29            "type": "object",
30            "properties": {
31                "query": {
32                    "type": "string",
33                    "description": "The search query"
34                }
35            },
36            "required": ["query"]
37        }
38    }
39]
40
41# Tool implementations
42def get_weather(city: str, unit: str = "celsius") -> dict:
43    """Simulated weather API."""
44    return {
45        "city": city,
46        "temperature": 22 if unit == "celsius" else 72,
47        "unit": unit,
48        "condition": "sunny"
49    }
50
51def search_web(query: str) -> dict:
52    """Simulated search."""
53    return {
54        "query": query,
55        "results": [
56            {"title": "Result 1", "url": "https://example.com/1"},
57            {"title": "Result 2", "url": "https://example.com/2"}
58        ]
59    }
60
61# Tool registry
62TOOL_FUNCTIONS = {
63    "get_weather": get_weather,
64    "search_web": search_web
65}
66
67def execute_tool(tool_name: str, arguments: dict) -> Any:
68    """Execute a tool by name with given arguments."""
69    if tool_name not in TOOL_FUNCTIONS:
70        raise ValueError(f"Unknown tool: {tool_name}")
71    
72    func = TOOL_FUNCTIONS[tool_name]
73    return func(**arguments)
74
75# Simulate tool calling flow
76print("Available tools:")
77for tool in TOOLS:
78    print(f"  • {tool['name']}: {tool['description']}")
79
80# Simulate LLM requesting a tool call
81tool_call = {
82    "name": "get_weather",
83    "arguments": {"city": "San Francisco", "unit": "celsius"}
84}
85
86print(f"\nTool call: {tool_call['name']}")
87print(f"Arguments: {tool_call['arguments']}")
88
89result = execute_tool(tool_call["name"], tool_call["arguments"])
90print(f"Result: {json.dumps(result, indent=2)}")