Structured outputs with structured memory

Because your Pydantic models deserve persistent context

The Problem

Instructor is magic. Define a Pydantic model, and the LLM returns perfectly structured data. No more parsing JSON strings. No more hoping the output matches your schema.

But here's the thing: Instructor helps you extract data. It doesn't help you remember it.

Every call is stateless. Every extraction starts fresh. The LLM doesn't know what schemas you've used before, what extraction patterns work best for your domain, or what edge cases you've already handled.

Your structured outputs are beautiful. Your institutional memory is nonexistent.

How Stompy Helps

Stompy gives Instructor the memory it deserves.

When you extract data, Stompy remembers: - Schema patterns that work for your domain - Extraction edge cases and how to handle them - Validation rules discovered over time - Retry strategies that actually succeed

Your extractions become smarter session by session—because they remember what worked before.

Integration Walkthrough

Combine Instructor with Stompy MCP

Use Instructor for structured outputs and Stompy for memory. They complement each other perfectly.

import instructor
from openai import OpenAI
from pydantic import BaseModel
# Stompy via your agent framework of choice

class UserProfile(BaseModel):
    name: str
    email: str
    preferences: dict

client = instructor.from_openai(OpenAI())

# Extract structured data
profile = client.chat.completions.create(
    model="gpt-4o",
    response_model=UserProfile,
    messages=[{"role": "user", "content": user_input}]
)

# Save extraction patterns to Stompy
# lock_context(topic="user_extraction_patterns",
#   content="UserProfile schema works well. Edge case: ...")