Comprehensive collection of tools adapted from OpenManus for smolagents CodeAct loop
# Install the base package
pip install smolagents-tools
# Install with specific feature sets
pip install smolagents-tools[search] # Web search capabilities
pip install smolagents-tools[browser] # Browser automation
pip install smolagents-tools[crawler] # Web crawling
pip install smolagents-tools[macos] # macOS automation (macOS only)
pip install smolagents-tools[llm] # LLM integrations
pip install smolagents-tools[all] # All features
# For development
pip install smolagents-tools[dev]from smolagents_tools import BashTool, WebSearchTool, PythonExecutorTool
from smolagents_tools import create_basic_toolset
# Use individual tools
bash_tool = BashTool()
result = bash_tool.forward(command="echo 'Hello World'")
# Use with smolagents CodeAgent
from smolagents import CodeAgent, InferenceClientModel
model = InferenceClientModel(model_id="Qwen/Qwen2.5-Coder-32B-Instruct")
tools = create_basic_toolset()
agent = CodeAgent(tools=tools, model=model, add_base_tools=True)
result = agent.run("Search for Python tutorials and create a summary")- BashTool - Command execution with proper error handling
- WebSearchTool - Multi-engine web search (DuckDuckGo, Google, Bing)
- PythonExecutorTool - Safe Python code execution with timeout
- FileEditorTool - Comprehensive file operations (view, create, edit, undo)
- BrowserTool - Web automation using Playwright
- WebCrawlerTool - Advanced web crawling with Crawl4AI
- MacOSUseTool - macOS application automation and UI interaction
- SimpleMacOSTool - Basic macOS operations and AppleScript execution
- PlanningTool - Task planning and management
- ChatCompletionTool - LLM integration (OpenAI, Anthropic, local)
- VNCComputerUseTool - VNC server automation and control
- SimpleVNCComputerUseTool - Basic VNC operations
- Basic Toolset (5 tools): Essential tools for most tasks
- Web Toolset (4 tools): Web-focused operations
- Development Toolset (7 tools): Software development workflow
- AI Toolset (5 tools): AI and language model integration
- VNC Toolset (2 tools): VNC desktop automation capabilities
Total: 23 available tools ready for use with smolagents!
# Test individual tools
from smolagents_tools import WebSearchTool
import asyncio
search_tool = WebSearchTool()
result = asyncio.run(search_tool.execute(query="AI news", max_results=5))
print(result.output)# Interactive mode with CodeAgent
from smolagents import CodeAgent, InferenceClientModel
from smolagents_tools import create_basic_toolset
model = InferenceClientModel(model_id="Qwen/Qwen2.5-Coder-32B-Instruct")
tools = create_basic_toolset()
agent = CodeAgent(tools=tools, model=model)
# Example tasks
agent.run("Search for the latest AI developments and summarize them")
agent.run("Create a Python script that analyzes data and saves results to a file")
agent.run("Research Python best practices and create a demo application")# Custom tool creation
from smolagents import tool
from smolagents_tools import WebSearchTool
@tool
def research_topic(topic: str) -> str:
"""Research a topic and provide insights"""
search_tool = WebSearchTool()
result = asyncio.run(search_tool.execute(query=topic, max_results=5))
return f"Research on {topic}:\n{result.output}"# Basic macOS operations
from smolagents_tools import create_tool
import asyncio
async def macos_example():
# Simple macOS tool for basic operations
simple_tool = create_tool("simple_macos")
# Text-to-speech
result = await simple_tool.execute(
action="say_text",
text="Hello from macOS automation!"
)
# Open applications
result = await simple_tool.execute(
action="open_app",
app_name="Calculator"
)
asyncio.run(macos_example())# VNC automation example
from smolagents_tools import create_tool
import asyncio
async def vnc_example():
# Create VNC tool
vnc_tool = create_tool("vnc_computer")
# Connect to VNC server
result = await vnc_tool.execute(
action="connect",
host="localhost",
port=5900
)
print(f"Connection result: {result}")
# Move mouse and click
result = await vnc_tool.execute(
action="mouse_click",
x=100,
y=100,
button=1
)
print(f"Mouse click result: {result}")
# Type text
result = await vnc_tool.execute(
action="type_text",
text="Hello VNC World!"
)
print(f"Type text result: {result}")
# Capture screen
result = await vnc_tool.execute(
action="capture_screen",
filename="screenshot.png"
)
print(f"Screen capture result: {result}")
asyncio.run(vnc_example())- Dynamic Forward Methods: Automatically generated method signatures
- Async-to-Sync Conversion: Seamless integration with CodeAgent
- Event Loop Handling: Robust nested event loop management
- Input Validation: Proper nullable parameter handling
- Command Execution: Bash commands with error handling
- Web Operations: Search, crawling, browser automation
- File Management: Create, edit, view files with undo support
- Code Execution: Safe Python execution with timeout
- AI Integration: Multiple LLM providers support
- Tool Registry: Easy tool discovery and instantiation
- Pre-configured Collections: Ready-to-use tool sets
- Comprehensive Testing: 100% test pass rate
- Rich Examples: Multiple usage patterns demonstrated
# Install in development mode
pip install -e '.[dev]'
# Run tests
python -m pytest
# Run the demo
smolagents-tools-demoagent.run("""
Search for recent developments in quantum computing,
analyze the key trends, and create a comprehensive report
with code examples demonstrating quantum concepts.
""")agent.run("""
Research modern web development best practices,
create a sample FastAPI application structure,
and include proper testing and documentation.
""")agent.run("""
Find datasets about climate change, create a Python script
to analyze the data, generate visualizations, and
summarize the key findings in a report.
""")agent.run("""
Create a script that monitors a website for changes,
sends notifications when updates are detected,
and logs all activities to a file.
""")For CodeAgent examples, you'll need API access:
# Hugging Face (Recommended)
export HUGGINGFACE_API_TOKEN="your_token_here"
# OpenAI (Alternative)
export OPENAI_API_KEY="your_openai_key_here"
# Or login via CLI
huggingface-cli loginWe welcome contributions! Here's how you can help:
- Add New Tools: Adapt more OpenManus tools
- Improve Examples: Create more usage demonstrations
- Fix Issues: Report and fix bugs
- Documentation: Improve guides and examples
To contribute to this library, first checkout the code. Then create a new virtual environment:
cd smolagents-tools
python -m venv venv
source venv/bin/activateNow install the dependencies and test dependencies:
python -m pip install -e '.[dev]'To run the tests:
python -m pytestThis project adapts tools from OpenManus for use with smolagents.
Licensed under the Apache License, Version 2.0.
- OpenManus: Original tool implementations
- Smolagents: Framework for agent-based automation
- Hugging Face: Infrastructure and model access
Ready to build powerful AI agents with comprehensive tool capabilities! 🚀