A powerful agent framework with enhanced capabilities including browser automation, code execution, MCP tool support, and deep research. Now defaults to EXTERNAL_MINION_AGENT framework for superior performance and functionality.
Minion Agent now defaults to the powerful EXTERNAL_MINION_AGENT framework, providing enhanced code execution, browser automation, and advanced planning capabilities out of the box.
pip install minion-agent-xgit clone [email protected]:femto/minion-agent.git
cd minion-agent
pip install -e .Here's a simple example of how to use Minion Agent:
import asyncio
import os
from dotenv import load_dotenv
from minion_agent import MinionAgent, AgentConfig, AgentFramework
from minion.agents import CodeAgent
import minion_agent
load_dotenv()
async def main():
# Configure the agent (using EXTERNAL_MINION_AGENT as default)
agent_config = AgentConfig(
model_id=os.environ.get("AZURE_DEPLOYMENT_NAME"),
name="research_assistant",
description="A helpful research assistant",
model_args={
"azure_endpoint": os.environ.get("AZURE_OPENAI_ENDPOINT"),
"api_key": os.environ.get("AZURE_OPENAI_API_KEY"),
"api_version": os.environ.get("OPENAI_API_VERSION"),
"model": "gpt-4o", # Actual model to use in minion framework
},
tools=[
minion_agent.tools.browser_tool.browser,
],
agent_type=CodeAgent, # Default agent type for EXTERNAL_MINION_AGENT
)
# Create agent with EXTERNAL_MINION_AGENT framework (default)
agent = await MinionAgent.create_async(AgentFramework.EXTERNAL_MINION_AGENT, agent_config)
# Run the agent with a question
result = await agent.run_async("What are the latest developments in AI?")
print("Agent's response:", result.final_output.content)
if __name__ == "__main__":
asyncio.run(main())see example.py see example_browser_use.py see example_with_managed_agents.py see example_deep_research.py see example_reason.py
Minion Agent supports multiple agent frameworks. The default framework is EXTERNAL_MINION_AGENT, which provides enhanced capabilities including:
- Code Generation and Execution: Advanced code generation with built-in execution capabilities
- Browser Automation: Integrated browser control and web interaction
- MCP Tool Support: Full Model Context Protocol integration
- Enhanced Planning: Sophisticated task planning and execution
EXTERNAL_MINION_AGENT(Default): Enhanced framework with code execution and browser capabilitiesSMOLAGENTS: HuggingFace's smolagents framework with planning supportLANGCHAIN: LangChain-based agentsOPENAI: OpenAI's assistant APIBROWSER_USE: Specialized for browser automation tasksDEEP_RESEARCH: Optimized for research and information gathering
When using EXTERNAL_MINION_AGENT, you can specify different agent types:
from minion.agents import CodeAgent, ToolCallingAgent
agent_config = AgentConfig(
# ... other config ...
agent_type=CodeAgent, # For code generation and execution
# or
# agent_type=ToolCallingAgent, # For general tool calling
)The AgentConfig class accepts the following parameters:
model_id: The ID of the model to use (e.g., "gpt-4")name: Name of the agent (default: "Minion")description: Optional description of the agentinstructions: Optional system instructions for the agenttools: List of tools the agent can usemodel_type: model type of the underlying agent frameworkmodel_args: Optional dictionary of model-specific argumentsagent_type: agent type of the underlying agent frameworkagent_args: Optional dictionary of agent-specific arguments
Minion Agent supports Model Context Protocol (MCP) tools. Here's how to use them:
from minion_agent.config import MCPTool
agent_config = AgentConfig(
# ... other config options ...
tools=[
minion_agent.tools.browser_tool.browser, # Regular tools
MCPTool(
command="npx",
args=["-y", "@modelcontextprotocol/server-filesystem", "/path/to/workspace"]
) # MCP tool
]
)You can also use MCP tools over Server-Sent Events (SSE). This is useful for connecting to remote MCP servers:
from minion_agent.config import MCPTool
agent_config = AgentConfig(
# ... other config options ...
tools=[
MCPTool({"url": "http://localhost:8000/sse"}), # SSE-based tool
]
)You can also use multiple MCP tools together:
tools=[
MCPTool(command="npx", args=["..."]), # Standard MCP tool
MCPTool({"url": "http://localhost:8000/sse"}), # SSE-based tool
MCPTool({"url": "http://localhost:8001/sse"}) # Another SSE-based tool
]You can enable automatic planning by setting the planning_interval in agent_args (smolagents) :
agent_config = AgentConfig(
# ... other config options ...
agent_args={
"planning_interval": 3, # Agent will create a plan every 3 steps
"additional_authorized_imports": "*"
}
)The planning_interval parameter determines how often the agent should create a new plan. When set to 3, the agent will:
- Create an initial plan for the task
- Execute 3 steps according to the plan
- Re-evaluate and create a new plan based on progress
- Repeat until the task is complete
Make sure to set up your environment variables in a .env file:
# For Azure OpenAI (recommended for EXTERNAL_MINION_AGENT)
AZURE_DEPLOYMENT_NAME=your_deployment_name
AZURE_OPENAI_ENDPOINT=https://your-resource.openai.azure.com/
AZURE_OPENAI_API_KEY=your_azure_api_key
OPENAI_API_VERSION=2024-02-15-preview
# Or for OpenAI
OPENAI_API_KEY=your_openai_api_key
# Optional: For other providers via LiteLLM
# ANTHROPIC_API_KEY=your_anthropic_key
# GOOGLE_API_KEY=your_google_keyTo set up for development:
# Clone the repository
git clone https://github.com/yourusername/minion-agent.git
cd minion-agent
# Create a virtual environment
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# Install development dependencies
pip install -e ".[dev]"See Deep Research Documentation for usage instructions.
Join our WeChat discussion group to connect with other users and get help:
群聊: minion-agent讨论群
MIT License