Trace Instructor structured outputs with Respan

Overview

Respan provides tracing capabilities for Instructor, enabling monitoring and analysis of structured LLM outputs through async workflows. Just 3 lines of setup, then your structured outputs are automatically traced.

The integration uses the @task decorator to wrap async functions that call Instructor. Define Pydantic models with field descriptions for structured output validation, and all tracing happens automatically.

How it works

Install instructor, openai, and keywordsai-tracing. Initialize KeywordsAITelemetry with your app name and desired instruments (OpenAI).

Set up an AsyncOpenAI client wrapped with instructor.from_openai(). Apply the @task decorator to async functions and run workflows normally - tracing happens in the background.

python

import instructor
from openai import AsyncOpenAI
from keywordsai_tracing import KeywordsAITelemetry, task
from pydantic import BaseModel

telemetry = KeywordsAITelemetry(app_name="my-app", instruments=["openai"])
client = instructor.from_openai(AsyncOpenAI())

class UserInfo(BaseModel):
    name: str
    age: int
    email: str

@task(name="extract_user")
async def extract(text: str) -> UserInfo:
    return await client.chat.completions.create(
        model="gpt-4o-mini",
        response_model=UserInfo,
        messages=[{"role": "user", "content": text}],
    )