Marvin

Structured data and tasks for Python applications

Marvin uses language models to turn unstructured input into Python values and to run tasks with tools. Pydantic AI handles model access and validates outputs.

import marvin

amount = marvin.cast("three dollars and fifty cents", float)
labels = marvin.classify("The service was excellent", ["positive", "negative"])

Start with a function when you need one result. Use an agent to reuse instructions and tools, a task to track an objective, and a thread to share conversation history.

Model outputs vary. Validating a result’s type does not establish that its content is correct.

Install Marvin, follow the quickstart, or browse the examples.