instructor

To extract structured JSON from LLM responses reliably, you validate the output against a Pydantic schema and automatically retry extraction when validation fails. This enforces strict types and constraints to produce trustworthy data for downstream pipelines.

Yes, Instructor supports multi-provider consistency for structured output across OpenAI and Anthropic APIs. It applies the same Pydantic-validated extraction pattern across both providers, utilizing JSON-mode fallbacks to ensure consistent results.

To handle validation errors when parsing complex JSON from LLMs, the system automatically feeds the validation error back to the model and re-attempts extraction. This retry mechanism ensures invalid data is corrected without manual intervention.

Yes, you can stream partial structured results while extracting data from an LLM. This feature allows you to receive real-time updates of Pydantic objects, enabling responsive user interfaces and incremental processing of data.

The best way to parse multiple entities from text using Pydantic is to define nested models with strict constraints and enums. The extraction process validates these schemas automatically and retries on failures to ensure accurate multi-entity parsing.

Yes, you need Pydantic model definitions to use structured output extraction. Defining these models with strict types, constraints, and nested structures is required to enforce validation and generate trustworthy JSON for downstream use.