Pydantic AI Integration

This integration is currently in beta. If you encounter any issues or unexpected behavior, please reach out for feedback and support.

Lunary supports Pydantic AI through OpenTelemetry instrumentation via Logfire.

To integrate Pydantic AI with Lunary, you only need to configure the OpenTelemetry exporter and instrument Pydantic AI:

os.environ["OTEL_EXPORTER_OTLP_ENDPOINT"] = "https://api.lunary.ai" # replace by your api endpoint if you're self-hosting Lunary
os.environ["OTEL_EXPORTER_OTLP_HEADERS"] = f"Authorization=Bearer {os.environ['LUNARY_PRIVATE_KEY']}"

logfire.configure(send_to_logfire=False)
logfire.instrument_pydantic_ai()

Full Example

Here’s a complete example showing how to use Pydantic AI with Lunary:

import os
import logfire
from pydantic import BaseModel
from pydantic_ai import Agent

os.environ["OTEL_EXPORTER_OTLP_ENDPOINT"] = "https://api.lunary.ai"
os.environ["OTEL_EXPORTER_OTLP_HEADERS"] = f"Authorization=Bearer {os.environ['LUNARY_PRIVATE_KEY']}"

logfire.configure(send_to_logfire=False)
logfire.instrument_pydantic_ai()


class MyModel(BaseModel):
    city: str
    country: str

agent = Agent(model='open:gpt-4.1', output_type=MyModel, model_settings={'temperature': 0.7})

if __name__ == '__main__':
    result = agent.run_sync('The windy city in the US of A.')
    print(result.output)

This will automatically track:

• Agent calls and responses
• Model parameters and settings
• Output schema validation
• Performance metrics
• Errors and exceptions

All telemetry data will be sent to Lunary for monitoring and analysis.