instructor

Use Instructor with a Pydantic BaseModel defining your output schema, then pass it as response_model to client.messages.create or client.chat.completions.create. Instructor validates the LLM output against your model and returns a typed Python object.

Instructor validates output with Pydantic and, on failure, sends the validation error back to the LLM and retries automatically. The default is 3 retries, configurable via the max_retries parameter.

Instructor provides full type safety, automatic Pydantic validation, automatic retries, and streaming with a low learning curve. LangChain offers partial type safety and no built-in validation retries, but suits complex chain orchestration.

Yes. Point an OpenAI client at the local Ollama server (base_url http://localhost:11434/v1) and wrap it with instructor.from_openai using Mode.JSON. You can then use response_model with models like llama3.1.

Yes. Use client.messages.create_partial to stream incremental updates of a Pydantic object, or create_iterable to stream list items one by one as the LLM generates them, enabling real-time UI updates.

You can use Field constraints like ge, le, min_length, max_length, and regex patterns, plus EmailStr and HttpUrl types, enums for fixed categories, custom field_validator functions, and model_validator for cross-field checks.