Skip to main content
Instructor is a Python library for generating structured outputs from LLMs using Pydantic models. Braintrust integrates with Instructor to trace structured output generation.

​
Setup

Install Instructor alongside the Braintrust SDK and OpenAI client: 
pip install braintrust instructor openai

​
Trace with Instructor

When using Instructor with Braintrust, wrap the OpenAI client with wrap_openai before patching with Instructor. This ensures Braintrust captures the low-level metrics and headers from OpenAI.
trace-instructor.py
import instructor
from braintrust import init_logger, wrap_openai
from openai import OpenAI
from pydantic import BaseModel

# Initialize Braintrust
logger = init_logger(project="Your project name")

# Define your response model
class MyResponseModel(BaseModel):
    name: str
    age: int

# Wrap OpenAI client with Braintrust FIRST, then patch with Instructor
client = instructor.patch(wrap_openai(OpenAI()))

# Use as normal - all calls are traced
response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Extract: John is 30 years old"}],
    response_model=MyResponseModel,
)
The order matters: instructor.patch(wrap_openai(OpenAI())) ensures Braintrust captures complete metrics.

​
Resources