Marvin

// Overview

Marvin is a Python framework for producing structured outputs from LLMs and building agentic AI workflows. It gives you a small set of functions like extract, cast, classify, and generate that turn unstructured text into typed Python objects, so model output flows back into your code as regular values you can validate and use.

It is aimed at Python developers who want type-safe results from language models without writing prompt scaffolding by hand. Marvin uses OpenAI by default but supports all Pydantic AI models, so you can point it at the provider you already use.

As an LLM orchestration tool, Marvin also offers a task-centric layer: you describe objectives as tasks, assign one or more agents to them, and compose tasks into threads to coordinate more complex, multi-step behavior.

// What it does

• Structured-output helpers: extract, cast, classify, and generate native Python types from any input
• Type-safe results validated against the result_type you specify (including TypedDict, Enum, and Pydantic types)
• Tasks as discrete, observable units of work that a default or named agent can run
• Named Agent objects with custom instructions for task-specific behavior
• Threads that compose tasks to orchestrate multi-step and multi-agent workflows
• Provider flexibility: defaults to OpenAI and supports all Pydantic AI models

// Getting started

Install Marvin, set your LLM provider key, then call one of its structured-output functions or run a task.

uv add marvin

export OPENAI_API_KEY=your-api-key

import marvin

result = marvin.extract(
    "i found $30 on the ground and bought 5 bagels for $10",
    int,
    instructions="only USD"
)
print(result)  # [30, 10]

import marvin

answer = marvin.run("the answer to the universe", result_type=int)
print(answer)  # 42

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Classify incoming text, such as routing support messages to the right department with a fixed set of labels
• Extract typed fields (numbers, dates, amounts) from messy or unstructured input for downstream code
• Cast free-form descriptions into structured objects like TypedDicts or Pydantic models
• Build multi-step agent workflows by composing tasks and assigning specialized agents

// How Marvin compares

Marvin alongside other open-source structured output tools AI/TLDR tracks, ranked by GitHub stars.