Instructor

Overview

Instructor patches the major LLM clients to return Pydantic models (or Zod schemas, in TS) instead of raw strings. It handles validation, retries, and streaming of structured output across providers. The first thing most engineering teams reach for when they need an LLM call that has to round-trip data reliably.

Best for

Best for
Engineers tired of regexing JSON out of model output

Use cases

• Force a model to return a well-typed JSON object on every call
• Add validation and retries to structured extraction tasks
• Stream typed structured output to the UI
• Use across many providers with one consistent API

Notes

Why it matters

Structured output is the unsung hero of production LLM apps. Instructor made the typed-object pattern the default, not the exception.

How teams use it in production

Patch the client, write a Pydantic model, ask for the model back. The rest of the code looks like a normal service. Reliability follows from the schema, not from clever prompt engineering.

What to watch

Native structured output is shipping across most major providers. Instructor’s value is the cross-provider consistency layer that stays useful when each provider does it slightly differently.