Pydantic AI

Pydantic AI is a Python agent framework that brings type-safety and ergonomic API design to generative AI development. The framework applies the same developer experience principles found in FastAPI to building production-grade AI applications.

This guide covers capturing telemetry from Pydantic AI applications using InteractiveAI.

Prerequisites

• InteractiveAI account with API credentials

• LLM provider credentials (OpenAI, Ollama, or other supported provider)

---

Installation

pip install interactiveai pydantic-ai

---

Configuration

Set your API credentials as environment variables:

import os

# InteractiveAI credentials
# Obtain keys from Settings > API Keys in the dashboard
os.environ["INTERACTIVEAI_PUBLIC_KEY"] = "pk-ia-..."
os.environ["INTERACTIVEAI_SECRET_KEY"] = "sk-ia-..."
os.environ["INTERACTIVEAI_HOST"] = "https://app.interactiveai.com"

# Model provider credentials
os.environ["OPENAI_API_KEY"] = "sk-proj-..."

Initialize the client and confirm connectivity:

from interactiveai import Interactive

client = Interactive(
    public_key=os.environ["INTERACTIVEAI_PUBLIC_KEY"],
    secret_key=os.environ["INTERACTIVEAI_SECRET_KEY"],
    host=os.environ.get("INTERACTIVEAI_HOST", "https://app.interactiveai.com")
)

if client.auth_check():
    print("Connection established")
else:
    print("Authentication failed - verify credentials")

---

Enabling Trace Capture

Pydantic AI includes built-in instrumentation. Activate it globally:

from pydantic_ai.agent import Agent

Agent.instrument_all()

This captures all agent operations and exports OpenTelemetry spans to InteractiveAI.

---

Running a Pydantic AI Agent

Here's a working example with a tool-equipped agent. Set instrument=True when configuring the agent:

from pydantic_ai import Agent, RunContext

quiz_agent = Agent(
    "openai:gpt-4o",
    deps_type=str,
    system_prompt=(
        "Use the `check_answer` function to verify if the "
        "user's answer matches the correct answer provided."
    ),
    instrument=True
)

@quiz_agent.tool
async def check_answer(ctx: RunContext[str], user_answer: str) -> str:
    """Check if the user's answer is correct"""
    correct = ctx.deps.lower()
    return "correct" if user_answer.lower() == correct else "incorrect"

# Run the agent
correct_answer = "paris"
result = await quiz_agent.run("I think the capital of France is Paris", deps=correct_answer)
print(result.output)

---

Enriching Traces with Context

Combine Pydantic AI instrumentation with the InteractiveAI SDK to attach identifiers and metadata:

from interactiveai import Interactive
from pydantic_ai import Agent, RunContext

client = Interactive(
    public_key=os.environ["INTERACTIVEAI_PUBLIC_KEY"],
    secret_key=os.environ["INTERACTIVEAI_SECRET_KEY"],
    host=os.environ.get("INTERACTIVEAI_HOST", "https://app.interactiveai.com")
)

quiz_agent = Agent(
    "openai:gpt-4o",
    deps_type=str,
    system_prompt="Verify answers using the check_answer tool.",
    instrument=True
)

@quiz_agent.tool
async def check_answer(ctx: RunContext[str], user_answer: str) -> str:
    correct = ctx.deps.lower()
    return "correct" if user_answer.lower() == correct else "incorrect"

async def run_with_context():
    with client.start_as_current_span(name="pydantic-ai-quiz") as span:
        
        client.update_current_trace(
            user_id="user",
            session_id="session",
            tags=["pydantic-ai", "quiz"],
            metadata={"category": "geography", "environment": "production"}
        )
        
        query = "The largest ocean is the Pacific"
        correct_answer = "pacific"
        result = await quiz_agent.run(query, deps=correct_answer)
        
        client.update_current_trace(
            input=query,
            output=result.output
        )
        
        return result

    client.flush()

result = await run_with_context()
print(result.output)

---

Trace Visibility

The InteractiveAI dashboard displays:

• Agent execution flow and tool invocations

• LLM calls with prompts and completions

• Tool function arguments and return values

• Token consumption and latency metrics

• Dependency injection context