567-labs/instructor ? reverse-engineered prompt

Build me a small Python library that makes it easy to get reliable structured JSON back from language models.

I want to define normal Pydantic models, send a prompt, and get back a validated Python object instead of messy text. It should work with a few major providers, especially OpenAI, but I’d like the same simple interface to work across providers if possible. Please handle the annoying parts for me, like parsing the response, checking it against the schema, retrying when validation fails, and supporting nested objects. If you can, add streaming so I can see partial results as they come in.

Keep the API really simple and friendly for everyday use, with clear examples and a few tests. Use current docs online if you need to.