Capability · Framework — orchestration

Instructor

Instructor is a focused library that makes structured outputs from LLMs a one-line change. You decorate an OpenAI-compatible client and pass a Pydantic response_model — Instructor handles function calling, JSON mode, validation, retries with re-asks on validation errors, and streaming of partial objects. It supports OpenAI, Anthropic, Gemini, Cohere, Ollama, LiteLLM and has ports in TypeScript, Go, Elixir and others.

Framework facts

Category: orchestration
Language: Python
License: MIT
Repository: https://github.com/567-labs/instructor

Install

pip install instructor

Quickstart

import instructor
from anthropic import Anthropic
from pydantic import BaseModel

class User(BaseModel):
    name: str
    age: int

client = instructor.from_anthropic(Anthropic())
user = client.messages.create(
    model='claude-opus-4-7',
    response_model=User,
    max_tokens=512,
    messages=[{'role': 'user', 'content': 'Ada is 36.'}]
)

Alternatives

Outlines — grammar-constrained generation
Pydantic AI — full agent framework with the same philosophy
Marvin — simpler decorators for common patterns
Native Structured Outputs (OpenAI/Anthropic)

Frequently asked questions

When should I use Instructor vs native structured outputs?

Native structured outputs (OpenAI Structured Outputs, Anthropic tool use) are great for single providers. Instructor unifies the API across providers, adds retry-on-validation-error, and handles streaming of partial objects — useful once you work with more than one model or care about resilience.

Does it work with Anthropic Claude?

Yes — via instructor.from_anthropic(). Instructor uses Anthropic tool use under the hood. It also supports Gemini, Cohere, Groq, Mistral, Ollama, and anything LiteLLM can reach.

Sources

Instructor — docs — accessed 2026-04-20
Instructor — GitHub — accessed 2026-04-20