An advanced AI agent system powered by smolagents for autonomous internet interactions, research, and intelligent task execution.
- 🔍 Web Search & Scraping: Search the web, scrape content, and gather information
- 🌍 Browser Use: Interact with websites, fill forms, click buttons, and extract structured data in the cloud using the Browser Use Cloud SDK
- 📰 News Analysis: Find and analyze recent news on any topic
- 💻 Code Execution: Execute Python code safely in a sandboxed environment
- 📊 Data Analysis: Analyze datasets and generate insights
- 🔬 Research Agent: Conduct multi-depth research on any topic
- 🛠️ Extensible Tools: Easy to add custom tools and capabilities
- 💬 Interactive Chat: Chat interface for continuous interaction
- 🎨 Rich CLI: Beautiful command-line interface with rich formatting
- 🚀 Multi-Model Orchestration: Strategic model assignment for specialized tasks
- 🌐 SerpAPI Integration: Advanced search with Google, Bing, Yahoo, Scholar, Maps, Shopping
- 🧠 Context Engineering: Intelligent context management and memory systems
- 📈 Cross-Engine Validation: Compare results across multiple search engines
The core process of the Agentic Internet is as follows:
# Clone the repository
git clone https://github.com/AgenticInternet/agentic-internet.git
cd agentic-internet
# Install with uv (recommended)
uv sync
# Or install with pip
pip install -e .Create a .env file in the project root:
# Optional: For HuggingFace models
HUGGINGFACE_TOKEN=your_huggingface_token
# Optional: For OpenAI models (GPT-4, etc.)
OPENAI_API_KEY=your_openai_api_key
# Optional: For Anthropic models (Claude)
ANTHROPIC_API_KEY=your_anthropic_api_key
# Optional: For SerpAPI advanced search
SERPAPI_API_KEY=your_serpapi_key
# Optional: For multi-model orchestration via OpenRouter
OPENROUTER_API_KEY=your_openrouter_key
# Optional: For Browser Use (Browser Use Cloud SDK)
BROWSER_USE_API_KEY=your_browser_use_keyfrom agentic_internet import InternetAgent
# Create an agent
agent = InternetAgent()
# Run a task
result = agent.run("Search for the latest AI news and summarize the top 3 stories")
print(result)
# Start interactive chat
agent.chat()uv run python -m agentic_internet.cli --help
Usage: python -m agentic_internet.cli [OPTIONS] COMMAND [ARGS]...
Advanced AI agent for intelligent internet interactions
╭─ Options ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ --help Show this message and exit. │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭─ Commands ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ chat Start an interactive chat session with the agent. │
│ run Run a single task with the agent. │
│ research Conduct research on a specific topic. │
│ config Manage configuration settings. │
│ multi Run a task using the multi-model orchestration system. │
│ orchestrate Execute complex tasks using model orchestration. │
│ tools Manage and display available tools. │
│ news Search and analyze news articles. │
│ models List and manage available AI models. │
│ version Display version information.
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
### Interactive Chat
```bash
# Start chat with default model
uv run python -m agentic_internet.cli chat --verbose
# Use a specific model
uv run python -m agentic_internet.cli run "Write a short story on GPT-5 models" --model gpt-5 --verbose
# Quiet mode (less verbose)
uv run python -m agentic_internet.cli run "Describe the latest AI breakthroughs" --model gpt-5 --verbose --quiet# Run a task
uv run python -m agentic_internet.cli run "Search for Python tutorials" --verbose
# Save output to file
uv run python -m agentic_internet.cli run "Analyze crypto market trends" --output results.txt --verbose
# Specify model and iterations
uv run python -m agentic_internet.cli run "Complex research task" --model claude-4 --max-iterations 15 --verbose# Quick research
uv run python -m agentic_internet.cli research "quantum computing" --depth quick
# Deep research with output
uv run python -m agentic_internet.cli research "AI ethics" --depth deep --output research.md
# Output as JSON
uv run python -m agentic_internet.cli research "climate change" --format json --output data.json# Multi-model task
uv run python -m agentic_internet.cli multi "Analyze the impact of AI on healthcare" --models gpt-5-chat --models claude-4 --workers 3 --output results.json
# News analysis
uv run python -m agentic_internet.cli news "artificial intelligence breakthroughs" --time 7d --sources "TechCrunch" --sources "Wired" --limit 20 --format markdown --output news_report.md# Orchestrated task
uv run python -m agentic_internet.cli orchestrate "Develop a business plan for an AI startup" --coordinator gpt-5-chat --workers claude-4 --output business_plan.json# List available tools
uv run python -m agentic_internet.cli tools --list
# Show configuration
uv run python -m agentic_internet.cli config --show
# Version info
uv run python -m agentic_internet.cli version# List all models
uv run python -m agentic_internet.cli models --list
# Show models by category
uv run python -m agentic_internet.cli models --category news
# Show detailed information
uv run python -m agentic_internet.cli models --detailsuv run python -m agentic_internet.cli toolsweb_search
Search the web for information using SerpAPI (if available) or DuckDuckGo.
Input should be a search query string.
Returns a list of search results with titles, snippets, and URLs.
web_scraper
Scrape and extract text content from a web page.
Input should be a URL to scrape.
Returns the extracted text content from the page.
news_search
Search for recent news articles on a topic.
Input should be a search query string.
Returns recent news articles with titles, descriptions, and sources.
browser_use
Advanced browser automation tool for interacting with websites, filling forms,
clicking buttons, and extracting structured data. Use this when you need to interact with
dynamic websites or perform complex web automation tasks.
async_browser_use
Async browser automation tool for high-performance web interactions.
Use this for complex scraping tasks that require parallel execution or streaming updates.
structured_browser_use
Browser automation with structured data extraction.
Use this when you need to extract specific structured information from websites.
python_executor
Execute Python code and return the output.
The code should be valid Python that can be executed.
Available libraries: numpy, pandas, requests, datetime, json, re, math.
Returns the output of the code execution or any errors.
data_analysis
Perform data analysis on CSV or JSON data.
Input should be a dictionary with 'data' (string of CSV/JSON) and 'operation' (what to analyze).
Returns analysis results including statistics, summaries, or visualizations.
- web_search: Search the web using DuckDuckGo
- web_scraper: Scrape and extract content from web pages
- news_search: Search for recent news articles
- multi_model_serpapi: Multi-model SerpAPI tools
- browser_use: Browser Use Cloud SDK
- python_executor: Execute Python code safely
- data_analysis: Analyze CSV/JSON data
- calculator: Perform mathematical calculations
uv run python -m agentic_internet.cli tools --multi╭────────── Tools ──────────╮
│ Multi-Model SerpAPI Tools │
╰───────────────────────────╯
GoogleSearchTool
Search the web using Google
GoogleShoppingTool
Search for products and prices
GoogleMapsLocalTool
Find local businesses and places
GoogleScholarTool
Search academic papers and citations
MultiEngineSearchTool
Search across multiple search enginesfrom agentic_internet import ResearchAgent
researcher = ResearchAgent()
# Conduct research at different depths
result = researcher.research(
topic="Artificial General Intelligence",
depth="deep" # Options: quick, moderate, deep
)
print(result['findings'])from agentic_internet import InternetAgent
agent = InternetAgent()
data = """
[
{"month": "Jan", "revenue": 10000, "costs": 6000},
{"month": "Feb", "revenue": 12000, "costs": 7000},
{"month": "Mar", "revenue": 15000, "costs": 8000}
]
"""
result = agent.run(f"Analyze this data and calculate profit margins: {data}")agent = InternetAgent(max_iterations=20)
complex_task = """
1. Search for the latest SpaceX launch
2. Find related news articles
3. Summarize the mission objectives
4. Analyze public sentiment from the news
"""
result = agent.run(complex_task)The system includes an advanced multi-model orchestration system that strategically assigns different models to specialized tasks:
import asyncio
from agentic_internet import MultiModelSerpAPISystem
# Initialize multi-model system
orchestrator = MultiModelSerpAPISystem(
serpapi_key="your_serpapi_key", # For advanced search
context_window_size=32768 # Large context window
)
# Setup specialized workers
orchestrator.setup_multi_model_workers()
# Execute complex workflow
async def run_analysis():
result = await orchestrator.execute_multi_model_workflow(
"Conduct comprehensive market analysis for AI coding assistants",
timeout=1200 # 20 minutes
)
return result
# Run the analysis
result = asyncio.run(run_analysis())- google_search: Advanced Google search with filters
- google_shopping: Product search with pricing
- google_maps_local: Local business search
- google_scholar: Academic paper search
- multi_engine_search: Cross-engine comparison (Google, Bing, Yahoo, Baidu)
- search_researcher: Multi-engine research specialist
- ecommerce_analyst: E-commerce and pricing analysis
- local_business_analyst: Local market intelligence
- academic_researcher: Academic research and citations
- competitive_analyst: Strategic competitive intelligence
- market_synthesizer: Comprehensive market synthesis
agentic-internet/
├── agentic_internet/
│ ├── __init__.py
│ ├── agents/ # Agent implementations
│ │ └── internet_agent.py
│ ├── tools/ # Custom tools
│ │ ├── web_search.py
│ │ └── code_execution.py
│ ├── config/ # Configuration
│ │ └── settings.py
│ ├── examples/ # Usage examples
│ │ └── basic_usage.py
│ └── cli.py # CLI interface
├── tests/ # Test suite
├── pyproject.toml # Project configuration
├── README.md # Documentation
└── .env # Environment variables
Configuration can be customized via environment variables or programmatically:
from agentic_internet import settings
# View current settings
print(settings.model.name)
print(settings.agent.max_iterations)
# Modify settings
settings.agent.verbose = False
settings.tools.web_search_enabled = TrueCreate custom tools by extending the Tool class:
from smolagents import Tool
class MyCustomTool(Tool):
name = "my_tool"
description = "Description of what the tool does"
inputs = {
"param1": {"type": "string", "description": "Parameter description"}
}
output_type = "string"
def forward(self, param1: str) -> str:
# Tool implementation
return f"Processed: {param1}"
# Use with agent
from agentic_internet import InternetAgent
agent = InternetAgent(tools=[MyCustomTool()])- HuggingFace Models: Llama, Mistral, etc. (via HuggingFace API)
- OpenAI: GPT-4, GPT-3.5-turbo (via OpenAI API)
- Anthropic: Claude 3 family (via Anthropic API)
- Local Models: Any model compatible with HuggingFace Transformers
uv run pytest tests/uv run black agentic_internet/
uv run ruff check agentic_internet/uv run mypy agentic_internet/MIT License - see LICENSE file for details
Contributions are welcome! Please feel free to submit a Pull Request.
- Built on top of smolagents by HuggingFace
- Uses DuckDuckGo Search for web searches
- Powered by various LLM providers (OpenAI, Anthropic, HuggingFace)
For questions and support, please open an issue on GitHub.