Instructor Structured LLM Outputs

Get reliable JSON from any LLM with Instructor. Generates Pydantic response models, validation+retry loops, streaming partial objects, and provider-agnostic clients for OpenAI, Anthropic, Mistral, and local models.

Stop writing regex to parse JSON out of model output. Instructor wraps any LLM client to return validated Pydantic objects, retrying with the validator error as context. This skill writes response schemas, streaming iterators, and patches custom retry strategies.

When to use

Use when you need structured data from LLMs — extraction pipelines, agent tool calls, form filling, classification at scale — and want type safety plus automatic retries when the model returns malformed output.

Examples

Entity extraction pipeline

Pull structured records out of free-form text

Use Instructor with Anthropic to extract Company, Role, and Salary fields from job-posting text into a typed Pydantic model

Streaming partial objects

Render fields to the UI as they arrive

Stream a partial Instructor response so I can render fields progressively in a Next.js app as the model generates them