First Agent

Agent, the one class you'll use constantly

Everything in Pydantic AI starts with an Agent. At minimum it needs a model string.

from pydantic_ai import Agent

agent = Agent("google:gemini-3.5-flash-lite")

The model string is "<provider>:<model-name>". google is the Gemini provider, reading GOOGLE_API_KEY from the environment exactly like ChatGoogleGenerativeAI did in the LangChain course, just without you having to import a separate provider class for the common case.

Running an agent

run_sync is the simplest way to call an agent: give it a prompt, get a result back.

from dotenv import load_dotenv
from pydantic_ai import Agent

load_dotenv()

agent = Agent("google:gemini-3.5-flash-lite")


def main() -> None:
    result = agent.run_sync("What is the capital of France? Answer in one sentence.")

    print("Output:", result.output)
    print("Output type:", type(result.output))

result isn't just the answer, it's an AgentRunResult object holding the output, the full message history, and usage stats. You'll use those other fields starting in Lesson 11. For now, result.output is all you need: with no output_type specified, it defaults to str, so this behaves exactly like model.invoke(...).content did in LangChain.

There's also an async run and a streaming run_stream (Lesson 9). run_sync is a thin wrapper that runs the async version for you, use it anywhere you're not already inside an async def.