mindchain 0.1.4

An agentic AI framework with Master Control Program (MCP) supervision

MindChain: Agentic AI Framework

MindChain is a comprehensive framework for building, deploying, and managing AI agents with a unique Master Control Program (MCP) supervision layer. The architecture enables both simple single-agent workflows and complex multi-agent systems with advanced coordination capabilities.

Installation

Install from PyPI:

pip install mindchain

Core Features

• Master Control Program (MCP): Central supervision system for agent management, monitoring, and orchestration
• Flexible Agent Architecture: Build single agents or complex multi-agent systems
• Memory Management: Contextual memory for improved agent responses
• Policy Enforcement: Control what agents can and cannot do
• Resource Monitoring: Track and limit resource usage
• Agent Orchestration: Coordinate multi-agent workflows with dependency management

Quick Start

import asyncio
from mindchain import MCP, Agent, AgentConfig

async def main():
    # Initialize the MCP
    mcp = MCP(config={
        'log_level': 'INFO',
        'policies': {
            'allow_external_tools': True,
        }
    })
    
    # Create agent configuration
    config = AgentConfig(
        name="ResearchAssistant",
        description="Helps with research tasks",
        system_prompt="You are a helpful research assistant that provides concise and accurate information."
    )
    
    # Create and register an agent
    agent = Agent(config)
    agent_id = mcp.register_agent(agent)
    
    # Run the agent with MCP supervision
    response = await mcp.supervise_execution(
        agent_id=agent_id,
        task=lambda: agent.run("What are the key benefits of transformer models?")
    )
    print(response)
    
    # Clean up
    mcp.unregister_agent(agent_id)

# Run in an async environment
if __name__ == "__main__":
    asyncio.run(main())

Multi-Agent Orchestration

MindChain provides a powerful AgentOrchestrator for coordinating complex multi-agent workflows:

import asyncio
from mindchain import MCP, Agent, AgentConfig, AgentOrchestrator

async def main():
    mcp = MCP()
    orchestrator = AgentOrchestrator(mcp)
    
    # Create and register multiple agents
    agents = []
    agent_ids = []
    
    for role in ["Researcher", "Analyst", "Writer"]:
        agent = Agent(AgentConfig(
            name=role,
            description=f"{role} agent",
            system_prompt=f"You are a {role.lower()} specialized in your domain."
        ))
        agent_id = mcp.register_agent(agent)
        agents.append(agent)
        agent_ids.append(agent_id)
    
    # Create a sequential workflow
    topic = "Artificial Intelligence in Healthcare"
    workflow_id = orchestrator.create_sequential_workflow(
        name="Research and Article Creation",
        description=f"Research and create an article about: {topic}",
        agent_ids=agent_ids,
        prompts=[
            f"Research the topic: {topic}",
            "Analyze this research: {previous_result}",
            "Write an article based on: {previous_result}"
        ]
    )
    
    # Execute the workflow
    results = await orchestrator.execute_workflow(workflow_id)
    
    # Get the final article from the last step
    workflow = orchestrator.get_workflow(workflow_id)
    final_step_id = workflow['steps'][-1]['id']
    article = results[final_step_id]
    
    print(article)
    
    # Clean up
    for agent_id in agent_ids:
        mcp.unregister_agent(agent_id)

if __name__ == "__main__":
    asyncio.run(main())

Framework Structure

• MCP: Centralized supervision layer
• Agent: Core agent implementation
• Memory: Remembers interactions for context
• Policies: Controls agent behaviors and permissions
• Resources: Manages and limits resource usage
• Orchestrator: Coordinates multi-agent workflows

Running Examples

Explore the included examples to see the framework in action:

# Run the simple agent example
python -m examples.basic_agent.simple_agent

# Run the multi-agent collaboration example
python -m examples.multi_agent.team_collaboration

Command Line Interface

MindChain comes with a built-in CLI:

# Show available commands
python -m mindchain --help

# Run an agent interactively
python -m mindchain run

# Run an agent with a specific query
python -m mindchain run --query "Explain what agentic AI is"

# Use a custom configuration file
python -m mindchain run --config my_agent_config.json

Testing

Run the tests:

pytest

Documentation

For detailed documentation, visit https://ali-beg.github.io/mindchain

Contributing

We welcome contributions! See CONTRIBUTING.md for details.

Roadmap

Check out our ROADMAP.md for upcoming features and development plans.

License

This project is licensed under the MIT License - see the LICENSE file for details.