LLM Models - Marvin

Marvin delegates model access to Pydantic AI. Use a provider:model-name string or a Pydantic AI model instance.

Set the default before importing Marvin

export OPENAI_API_KEY="your-api-key"
export MARVIN_AGENT_MODEL="openai:gpt-4o-mini"

Provider credentials use the provider’s environment variables. Marvin-specific settings use the MARVIN_ prefix.

Choose a model for an agent

import marvin

agent = marvin.Agent(model="openai:gpt-4o-mini")
answer = agent.run("What is 23 + 19?", result_type=int)

For Anthropic, export ANTHROPIC_API_KEY and use an Anthropic model identifier, such as anthropic:claude-haiku-4-5.

Change the application default

import marvin

marvin.defaults.model = "openai:gpt-4o-mini"

For a temporary override:

import marvin
from marvin.defaults import override_defaults

with override_defaults(model="openai:gpt-4o-mini"):
    answer = marvin.run("What is 23 + 19?", result_type=int)

Pass provider-supported settings through Agent(model_settings={...}). Accepted settings and model identifiers depend on the provider; consult its Pydantic AI documentation rather than assuming that all models accept temperature or the same options.

⌘I