Tool-calling agents

Let the model call real Python functions to answer user queries.

This example shows how to create agents that can call real Python functions. We’ll demonstrate both the low-level approach and the preferred Agent class approach.

1. Define the tool

   from ai_sdk import tool
   
   add = tool(
       name="add",
       description="Add two numbers and return the sum.",
       parameters={
           "type": "object",
           "properties": {
               "a": {"type": "integer"},
               "b": {"type": "integer"},
           },
           "required": ["a", "b"],
       },
       execute=lambda a, b: a + b,
   )
   

2. Using the Agent class (Recommended)

   The Agent class provides a more convenient interface with persistent state and better organization:

   from ai_sdk import Agent, openai
   from tools import add
   
   model = openai("gpt-4.1-mini", temperature=0)
   agent = Agent(
       name="Math Assistant",
       model=model,
       system="You are a helpful math assistant.",
       tools=[add]
   )
   
   response = agent.run("What is 21 + 21?")
   print(response)  # "The result is 42."
   

   The Agent class is the recommended approach for most use cases. It provides persistent state, better error handling, and a cleaner interface for multi-turn conversations.

3. Low-level approach (Advanced)

   For advanced use cases, you can use the lower-level generate_text function directly:

   from ai_sdk import openai, generate_text
   from tools import add  # assume above code lives in tools.py
   
   model = openai("gpt-4.1-mini", temperature=0)
   
   res = generate_text(
       model=model,
       prompt="What is 21 + 21?",
       tools=[add],
   )
   print(res.text) # "The result is 42."
   print(res.tool_calls) # introspection
   print(res.tool_results)
   

The SDK automatically loops until the model stops requesting tools (max 8 iterations by default). Each tool result is appended as a tool message so the model can reference previous calls.