interpreter-smol: Installation and Usage Guide

This guide will help you set up and use interpreter-smol, a thin wrapper around SmolaGents that provides an Open-Interpreter-like experience with Gemini support.

Installation

1. Install the required packages:

# Install SmolaGents and base requirements
pip install smolagents

# Install Gemini support (required for default model)
pip install google-genai

2. Download and install the interpreter-smol package:

# Clone the repository
git clone https://github.com/your-username/interpreter-smol.git
cd interpreter-smol

# Install the package
pip install -e .

3. Set up your API keys as environment variables:

# For Gemini (recommended)
export GOOGLE_API_KEY="your-api-key-here"

# For other providers (optional)
export OPENAI_API_KEY="your-api-key-here"
export ANTHROPIC_API_KEY="your-api-key-here"
export HF_API_TOKEN="your-api-key-here"

Basic Usage

Command Line Interface

The command line interface is designed to be simple and familiar to Open-Interpreter users:

# Basic usage (defaults to Gemini model)
interpreter-smol "Write a Python function to calculate the Fibonacci sequence"

# Start an interactive chat session
interpreter-smol

# Specify a different model provider
interpreter-smol --model openai "Write a function to calculate prime numbers"

# Enable specific tools
interpreter-smol --tools unrestricted_python web_search "Find the current Bitcoin price and plot it"

# Allow additional Python imports
interpreter-smol --imports numpy pandas matplotlib.pyplot scipy "Create a visualization of the sine function"

# Enable verbose output
interpreter-smol -v "Calculate the factorial of 10"

Python API

You can also use the interpreter programmatically in your Python code:

from interpreter_smol import Interpreter

# Initialize with default settings (Gemini)
interpreter = Interpreter()

# Run a single prompt
result = interpreter.run("Write a Python function to calculate prime numbers")

# Start an interactive chat session
interpreter.chat("Help me analyze this dataset")

# Customize model and tools
custom_interpreter = Interpreter(
    model="anthropic",
    model_id="claude-3-5-sonnet-20240620",
    tools=["unrestricted_python", "web_search", "visit_webpage"],
    imports=["numpy", "pandas", "scipy", "matplotlib.pyplot"],
    verbose=True
)
custom_interpreter.run("Analyze recent stock market trends")

Available Tools

The following tools are available:

• unrestricted_python: Execute Python code
• web_search: Search the web with DuckDuckGo
• visit_webpage: Visit and extract content from a webpage

Supported Models

• Gemini (default): Google's Gemini 2.0 models
• OpenAI: GPT-3.5, GPT-4 models
• Anthropic: Claude models
• Hugging Face: Models hosted on Hugging Face

Configuration Options

Option	Description	Default
--model, -m	Model provider	gemini
--model-id	Specific model identifier	Depends on provider
--api-key, -k	API key	Uses environment variable
--tools	Tools to enable	unrestricted_python, web_search
--imports	Additional Python modules to allow	numpy, pandas, matplotlib.pyplot
--temperature, -t	Temperature for generation	0.7
--max-tokens	Maximum tokens in response	4096
-i, --interactive	Start in interactive mode	False
-v, --verbose	Enable verbose output	False

Advanced Usage with SmolaGents

For more advanced use cases, you can access the SmolaGents functionality directly:

from smolagents import CodeAgent
from smolagents import LiteLLMModel
from smolagents.default_tools import TOOL_MAPPING

# Create a custom Gemini model
model = LiteLLMModel(model_id="gemini/gemini-2.0-flash")

# Create an agent with specific tools
agent = CodeAgent(
    tools=[TOOL_MAPPING["unrestricted_python"](), TOOL_MAPPING["web_search"]()],
    model=model,
    additional_authorized_imports=["numpy", "pandas"],
    verbosity_level=2
)

# Run the agent
agent.run("Your complex task here")

Differences from Open-Interpreter

interpreter-smol offers several advantages compared to Open-Interpreter:

1. Native Gemini 2.0 support: First-class integration with Google's latest Gemini models
2. Modern agent framework: Built on SmolaGents, providing a more robust foundation
3. Enhanced security: Better code execution security options
4. Simplified implementation: Thin wrapper focused on essential functionality
5. Open-source foundation: Built on the well-maintained SmolaGents library

Limitations

• Some advanced features from Open-Interpreter (like vision) require additional setup
• Currently focused on core functionality; some specialized features may not be available yet