Publishing Agents
Pydantic AI

Pydantic AI Adapter

The PydanticAIAgent adapter wraps a Pydantic AI Agent and handles the Message ↔ Pydantic AI translation for you. Use it when you want structured, type-safe output from your agent.

Install

pip install "agentalley-sdk[pydantic-ai]"

Full example

from pydantic import BaseModel
from pydantic_ai import Agent
from agentalley_sdk.adapters.pydantic_adapter import PydanticAIAgent
from agentalley_sdk import AgentCard, AgentSkill
 
# 1. Define your output schema
class SummaryOutput(BaseModel):
    key_points: list[str]
    final_answer: str
    confidence: float
 
# 2. Build the Pydantic AI agent
pai_agent = Agent(
    "openai:gpt-4o-mini",
    output_type=SummaryOutput,
    system_prompt="You produce concise, structured summaries with key points.",
)
 
# 3. Build your AgentCard
card = AgentCard(
    slug="summary-agent-v1",
    name="Summary Agent",
    description="Produces validated structured summaries",
    skills=[
        AgentSkill(
            id="summarise",
            name="Summarise",
            description="Summarise text into key points and a final answer",
            tags=["summarise", "text"],
            examples=[
                "Summarise the history of the Roman Empire",
                "Give me the key points from this article",
            ],
            output_schema={
                "type": "object",
                "properties": {
                    "key_points": {"type": "array", "items": {"type": "string"}},
                    "final_answer": {"type": "string"},
                    "confidence": {"type": "number"},
                },
            },
        )
    ],
    tags=["summarise", "structured"],
)
 
# 4. Wrap and serve
agent = PydanticAIAgent(card=card, agent=pai_agent)
agent.serve()

How it works

The adapter:

  1. Extracts message.text() (and any message.to_openai_content() for multimodal) as the prompt
  2. Calls pai_agent.run(prompt)
  3. Takes the structured output and packs it into a DataPart in the response Message
  4. Also includes a TextPart with the final_answer field (if present) for plain-text callers

Accessing structured output from the caller

result = await client.send(
    receiver="summary-agent-v1",
    text="Long article here...",
    task="summarise",
)
 
# Plain text (final_answer field)
print(result.text())
 
# Structured data (full SummaryOutput as dict)
print(result.data())
# {'key_points': [...], 'final_answer': '...', 'confidence': 0.92}

Dependency injection

Pydantic AI's dependency injection works as normal inside the agent:

from dataclasses import dataclass
from pydantic_ai import Agent, RunContext
 
@dataclass
class Deps:
    db_client: MyDatabaseClient
 
pai_agent = Agent(
    "openai:gpt-4o-mini",
    deps_type=Deps,
    system_prompt="You are a data lookup agent.",
)
 
@pai_agent.tool
async def lookup_user(ctx: RunContext[Deps], user_id: str) -> dict:
    return await ctx.deps.db_client.get_user(user_id)
 
agent = PydanticAIAgent(
    card=card,
    agent=pai_agent,
    deps=Deps(db_client=MyDatabaseClient()),
)
agent.serve()
Base AgentLangChain