Yes. Marvin 3 uses Pydantic AI under the hood and works with any OpenAI‑compatible endpoint, including Ollama. Quickstart (local Ollama) 1) Start Ollama (defaults to http://localhost:11434) 2) Set the OpenAI‑compatible base URL (note the /v1 suffix):

export OLLAMA_BASE_URL="<http://localhost:11434/v1>"

3) Use an Ollama model in Marvin:

import marvin

agent = marvin.Agent(model="ollama:llama3.2")  # e.g., llama3.2, mistral, qwen2.5, etc.
print(agent.run("Write a haiku about Prefect"))

Explicit provider setup (if you want to be precise in code)

import marvin
from pydantic_ai.models.openai import OpenAIChatModel
from pydantic_ai.providers.ollama import OllamaProvider

ollama = OpenAIChatModel(
    "llama3.2",
    provider=OllamaProvider(base_url="<http://localhost:11434/v1>")
)

agent = marvin.Agent(model=ollama)
print(agent.run("What's 2+2?"))

Notes - Use any Ollama model name you’ve pulled (e.g., llama3.2, mistral, deepseek-r1). - If you see connection errors, double‑check that the base URL includes /v1 and that Ollama is running. - You can also route through a LiteLLM proxy if you prefer. Links - Marvin 3 (GitHub) - Pydantic AI – OpenAI (Ollama‑compatible) model - Ollama If you share your target model and environment (local vs. remote), I can tailor the config.