Pydantic AI: AI Agent Library. AI agents is an evolving landscape in… | by Naresh Kancharla | Dec, 2024

AI brokers is an evolving panorama in AI Improvement and discovering a proper device which adapts and open sufficient for the longer term necessities makes a distinction. Pydantic AI library is highly effective AI agent library constructed by the crew behind the favored Pydantic information validation library

Pydantic AI is an agentic framework designed to simplify the method of making and managing AI brokers. It builds upon the strong basis of the Pydantic library, which is broadly used for information validation and kind checking in Python initiatives.

pip set up pydantic-ai

If you already know which mannequin you’re going to make use of and need to keep away from putting in superfluous packages, you should utilize the pydantic-ai-slim bundle.

pip set up pydantic-ai-slim 

Creating an agent is easy, requiring only a few strains of code. You may outline an agent by specifying the mannequin identify and a system immediate.

from pydantic_ai import Agent
agent = Agent(  
'ollama:llama3'
#system_prompt='Be concise, reply with one sentence.',  
)
consequence = agent.run_sync('The place have been the olympics held in 2012 ?')  
print(consequence.information)
"""
The Olympics held in 2012 have been the Summer season Olympics, formally often known as the 
Video games of the XXX Olympiad. They occurred in London, United Kingdom, 
from July 27 to August 12, 2012. The Paralympic Video games adopted 
from August 29 to September 9, 2012, additionally in London.
"""

Harnesses the facility of Pydantic to validate and structure mannequin outputs, making certain responses are constant throughout runs.

from pydantic import BaseModel
from pydantic_ai import Agent
class CityLocation(BaseModel):
metropolis: str
nation: str
agent = Agent('ollama:llama3.2', result_type=CityLocation)
consequence = agent.run_sync('Which metropolis and nation have been the olympics held in 2012?')
print(consequence.information)
"""

metropolis='London' nation='United Kingdom'
"""

Operate instruments present a mechanism for fashions to retrieve further info to assist them generate a response.

There are a variety of the way to register instruments with an agent:

• by way of the @agent.tool decorator — for instruments that want entry to the agent context
• by way of the @agent.tool_plain decorator — for instruments that don’t want entry to the agent context
• by way of the tools key phrase argument to Agent which might take both plain capabilities, or situations of Tool

@agent.device is taken into account the default decorator since within the majority of instances instruments will want entry to the agent context.

import random
from pydantic_ai import Agent, RunContext
agent = Agent(
'ollama:llama3.2',  
deps_type=str,  
system_prompt=(
"You are a cube sport, you need to roll the die and see if the quantity "
"you get again matches the person's guess. If that's the case, inform them they are a winner. "
"Use the participant's identify within the response."
),
)
@agent.tool_plain  
def roll_die() -> str:
"""Roll a six-sided die and return the consequence."""
return str(random.randint(1, 6))
@agent.device  
def get_player_name(ctx: RunContext[str]) -> str:
"""Get the participant's identify."""
return ctx.deps
dice_result = agent.run_sync('My guess is 4', deps='Anne')  
print(dice_result.information)