instructor

To get structured JSON output from an LLM with Pydantic validation, you define a Pydantic model and pass it to the extraction client. The tool parses the LLM response and validates it against your schema, automatically requesting corrections if validation fails.

Yes, you can stream partial Pydantic objects during LLM generation. The tool emits iterable partial objects as the model generates text, enabling real-time UI updates and progressive data display without waiting for the full response.

Yes, this structured output tool works with both OpenAI and Anthropic models. It supports provider-specific JSON modes and local model deployments, allowing entity extraction and batch processing across multiple LLM providers.

The best way to extract entities from large document corpora into analytics pipelines is using schema-validated parsing with batch processing. This approach normalizes nested and union types, guaranteeing schema conformance for downstream analytics ingestion.

LLM output fails JSON parsing when handling nested and union types due to inconsistent free-form responses. This tool solves the problem by applying robust JSON parsing and type safety, normalizing values and performing model-level checks for complex business logic.

Automatic retry works by re-prompting the model for corrected output when validation fails. If the initial LLM response does not match the Pydantic schema, the tool automatically requests a corrected response to ensure strict schema conformance.