Feature: Memory-backed extraction — recall past structured outputs for few-shot context

Instructor extracts structured data from LLMs precisely, but every extraction is independent — no shared memory means repeated similar extractions re-do the same work. This proposes Dakera (https://dakera.ai) as an optional memory layer for memory-backed few-shot extraction.

Problem: When using Instructor to extract the same class of structured data repeatedly, high-quality past extractions are never used as few-shot examples. Each extraction starts from scratch even if similar data was extracted successfully yesterday.

Proposed: memory= parameter on client.chat.completions.create:

import instructor from openai import OpenAI from pydantic import BaseModel from dakera import DakeraClient

class InvoiceData(BaseModel): vendor: str total: float date: str

memory = DakeraClient(base_url='http://localhost:3300', api_key='demo') client = instructor.from_openai(OpenAI())

def extract_with_memory(text: str, agent_id: str = 'invoice-extractor') -> InvoiceData: # Recall prior successful extractions as few-shot context prior = memory.recall(agent_id=agent_id, query=text[:200], top_k=2) system = 'Extract structured invoice data.' if prior and prior.memories: examples = '\n'.join(f'Example: {m.content}' for m in prior.memories) system = f'{system}\n\nPrior successful extractions:\n{examples}'

result = client.chat.completions.create(
    model='gpt-4o-mini',
    messages=[{'role': 'system', 'content': system}, {'role': 'user', 'content': text}],
    response_model=InvoiceData,
)

# Store successful extraction for future recall
memory.store_memory(agent_id=agent_id, content=result.model_dump_json(), metadata={'source': text[:100]})
return result

Setup: docker run -d -p 3300:3300 -e DAKERA_API_KEY=demo ghcr.io/dakera-ai/dakera:latest / pip install dakera

This could be added as a cookbook recipe or optional InstructorWithMemory wrapper.

Happy to open a PR.