LLM Models - Marvin

Marvin uses Pydantic AI as its underlying LLM engine, providing a clean, type-safe interface for working with various LLM providers.

The Default Model

By default, Marvin uses OpenAI’s GPT-4o model. To use this default configuration, you only need to set your OpenAI API key:

export OPENAI_API_KEY="your-api-key"

Configuring Models

Using String Identifiers

The simplest way to specify a model is with a string identifier in the format provider:model_name:

import marvin

# Create an agent with a specific model
agent = marvin.Agent(model="openai:gpt-4o-mini")

Model Settings

You can customize model behavior using the model_settings parameter when creating an agent:

import marvin

agent = marvin.Agent(
    name="Creative Writer",
    model="openai:gpt-4o",
    model_settings={
        "temperature": 0.8,

    }
)

Changing the Default Model

You can change the default model for all agents in your application:

import marvin

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

result = marvin.run("Write a haiku")

Environment Variables

You can also configure the default model and other settings using environment variables:

# Set the default model
export MARVIN_AGENT_MODEL="openai:gpt-4"

# Configure model settings
export MARVIN_AGENT_TEMPERATURE=0.7

Model Providers

Marvin supports any model provider that is compatible with Pydantic AI. Common providers include:

OpenAI
Anthropic
Azure OpenAI
Google

Each provider may require its own API key and configuration. Refer to the provider’s documentation for specific setup instructions.

Passing Configuration to Models

For more control, you may pass pydantic-ai models when creating an agent. For example, to use Anthropic’s Claude 3.5 Sonnet model with a custom httpx client:

export ANTHROPIC_API_KEY="your-api-key"

import os
from pathlib import Path

import httpx
from pydantic_ai.models.anthropic import AnthropicModel
import marvin

def write_file(path: str, content: str):
    """Write content to a file"""
    _path = Path(path)
    _path.write_text(content)

writer = marvin.Agent(
    model=AnthropicModel(
        "anthropic/claude-3-5-sonnet@latest",
        http_client=httpx.AsyncClient(
            timeout=10,
            # proxy="http://localhost:8080",
            headers={"x-api-key": os.getenv("ANTHROPIC_API_KEY", "gonna fail")},
        ),
    ),
    name="Technical Writer",
    instructions="Write concise, engaging content for developers",
    tools=[write_file],
)

result = marvin.run("how to use pydantic? write haiku to docs.md", agents=[writer])
print(result)

​The Default Model

​Configuring Models

​Using String Identifiers

​Model Settings

​Changing the Default Model

​Environment Variables

​Model Providers

​Passing Configuration to Models