masai-framework 0.4.7

Multi-Agent System Framework for AI Agents with Vanilla SDK Wrappers and Custom LangGraph

MASAI - Multi-Agent System AI Framework

A powerful, production-ready framework for building multi-agent AI systems with advanced features like persistent memory, long-context management, and sophisticated agent orchestration.

⭐ Please star this project if you find it useful!

🆕 New Documentation

We've added comprehensive guides to help you get the most out of MASAI:

• Model Parameters Guide - Complete reference for all supported models (Gemini, OpenAI, Anthropic) with ALL parameters, examples, and best practices
• Tools Guide - How to define tools, use them, implement Redis caching, and integrate with agents
• Singular Agent Guide - Complete guide for single agent architecture, execution, memory, and tools
• Multi-Agent System Guide - Comprehensive guide for decentralized and hierarchical multi-agent coordination
• OMAN Guide - Orchestrated Multi-Agent Network for enterprise-level multi-domain systems

---

📋 Quick Navigation

Getting Started

Document	Description
Quick Start	Get started in 5 minutes
Installation	Setup instructions and requirements
Configuration	Configuration options and setup

Core Concepts

Document	Description
Framework Overview	Architecture and design principles
Model Parameters	NEW! Complete model configuration guide
Tools Guide	NEW! Tool definition, usage, and caching
Memory System	Persistent memory and long-context management

Agent Systems

Document	Description
Agent Manager Detailed	AgentManager API and usage
Singular Agent Guide	NEW! Single agent architecture and usage
Multi-Agent System Guide	NEW! Decentralized and hierarchical MAS
OMAN Guide	NEW! Orchestrated Multi-Agent Network

Advanced Topics

Document	Description
Advanced Usage	Expert patterns and techniques
Multi-Agent Orchestration	Complex multi-agent workflows
LangChain Agnostic Guide	Using MASAI without LangChain

Reference

Document	Description
API Reference	Complete API documentation
Troubleshooting	Common issues and solutions
Usage Guide	Common usage patterns

---

Quick Start

Installation

pip install masai-framework

Basic Usage

from masai.AgentManager import AgentManager, AgentDetails
import asyncio

# Create agent manager
manager = AgentManager(user_id="user_123")

# Create an agent
agent = manager.create_agent(
    agent_name="assistant",
    tools=[],  # Add LangChain tools here
    agent_details=AgentDetails(
        capabilities=["analysis", "reasoning"],
        description="Helpful assistant",
        style="concise"
    )
)

# Use the agent - Full execution
result = await agent.initiate_agent(
    query="What is 2+2?",
    passed_from="user"
)
print(result["answer"])

# Or use streaming for real-time updates
async for state in agent.initiate_agent_astream(
    query="What is 2+2?",
    passed_from="user"
):
    node_name, state_dict = state
    state_value = [v for k, v in state_dict.items()][0]
    print(f"Node: {state_value['current_node']}")

See docs/QUICK_START.md for detailed examples.

---

Core Features

🧠 Multi-Agent Architecture

• Router-Evaluator-Reflector Pattern: Sophisticated agent decision-making
• Agent Orchestration: Coordinate multiple agents for complex tasks
• Tool Integration: Seamless integration with LangChain tools
• Streaming Support: Real-time response streaming

💾 Persistent Memory

• Redis Backend: Fast vector storage with RediSearch
• Qdrant Backend: Distributed vector database support
• User Isolation: Multi-user support with automatic filtering
• Deduplication: Automatic duplicate detection and merging

🔄 Long-Context Management

• Context Summarization: Automatic summarization of long conversations
• Memory Overflow Handling: Intelligent flushing to persistent storage
• Semantic Search: Find relevant memories using embeddings
• Category Filtering: Organize memories by categories

🎯 Flexible Configuration

• Multiple LLM Providers: OpenAI, Google Gemini, Anthropic Claude
• Custom Embeddings: Support for any embedding model
• Scalable Parameters: Configure all model parameters via config
• Component Customization: Override any component behavior

🤝 Multi Agent Orchestration

• Sequential Workflow: Fixed agent pipeline
• Hierarchical Workflow: Supervisor-based delegation
• Decentralized Workflow: Peer-to-peer collaboration
• Orchestrated Multi-Agent Network (OMAN): Coordinate multiple MAS instances
• Data & Context Management: Shared memory, context propagation, isolation
• See docs/MULTI_AGENT_ORCHESTRATION.md for details

---

📚 Documentation

Core Documentation

• Quick Start - Get started in 5 minutes
• Architecture - System design and components
• Memory System - Persistent memory and long-context management
• Installation - Complete setup guide
• Configuration - All configuration options

Advanced Documentation

• Multi-Agent Orchestration - Sequential, hierarchical, decentralized, and OMAN patterns
• Data & Context Management - Data flow, context sharing, and isolation
• Usage Guide - Common usage patterns
• Advanced Topics - Expert patterns and customization
• Troubleshooting - Common issues and solutions
• API Reference - Complete API documentation

---

Architecture

System Overview

User Application
       ↓
AgentManager (Orchestrator)
       ↓
Agent (Router-Evaluator-Reflector)
       ├─ MASGenerativeModel (LLM + Memory)
       ├─ Tool Executor
       └─ State Manager
       ↓
Memory System
       ├─ LongTermMemory
       ├─ Redis/Qdrant Backend
       └─ Embedding Model

See docs/ARCHITECTURE.md for detailed architecture.

---

Installation

Requirements

• Python 3.8+
• Redis (for persistent memory) or Qdrant
• API keys for LLM providers (OpenAI, Google, etc.)

Setup

# Clone repository
git clone https://github.com/shaunthecomputerscientist/mas-ai.git
cd mas-ai

# Install dependencies
pip install -r requirements.txt

# Set up environment variables
cp .env.example .env
# Edit .env with your API keys

See docs/INSTALLATION.md for detailed setup.

---

Usage Guide

Creating Agents

agent = manager.create_agent(
    agent_name="research_agent",
    agent_details=AgentDetails(
        capabilities=["research", "analysis"],
        description="Research specialist",
        style="detailed"
    ),
    tools=[]  # Add LangChain tools here
)

Executing Agent

# Full execution
result = await agent.initiate_agent(
    query="Explain quantum computing",
    passed_from="user"
)
print(result["answer"])
print(f"Reasoning: {result['reasoning']}")
print(f"Satisfied: {result['satisfied']}")

Streaming Responses

async for state in agent.initiate_agent_astream(
    query="Tell me a story",
    passed_from="user"
):
    node_name, state_dict = state
    state_value = [v for k, v in state_dict.items()][0]

    # Access state information
    if state_value.get("answer"):
        print(state_value["answer"])

See docs/USAGE_GUIDE.md for comprehensive examples.

---

Memory System

Persistent Memory Setup

from masai.AgentManager import AgentManager
from masai.Memory.LongTermMemory import RedisConfig
from langchain_openai import OpenAIEmbeddings

# Configure memory backend
redis_config = RedisConfig(
    redis_url="redis://localhost:6379",
    index_name="masai_vectors",
    vector_size=1536,
    embedding_model=OpenAIEmbeddings(model="text-embedding-3-small")
)

# Create manager with memory config
manager = AgentManager(
    user_id="user_123",
    model_config_path="model_config.json",
    memory_config=redis_config  # Pass to AgentManager
)

# Create agent with persistent memory enabled
agent = manager.create_agent(
    agent_name="assistant",
    agent_details=AgentDetails(
        capabilities=["reasoning"],
        description="Assistant"
    ),
    persist_memory=True,  # Enable persistence
    long_context=True,
    long_context_order=5
)

Memory Operations

# Access long-term memory through agent's LLM components
from masai.schema.Document import Document

# Save memories
await agent.llm_router.long_term_memory.save(
    user_id="user_123",
    documents=[Document(page_content="User likes Python")]
)

# Search memories
memories = await agent.llm_router.long_term_memory.search(
    user_id="user_123",
    query="What does user like?",
    k=5
)

See docs/MEMORY_SYSTEM.md for detailed memory docs.

---

Configuration

Agent Configuration Parameters

Parameter	Type	Default	Description
agent_name	str	Required	Unique agent identifier
agent_details	AgentDetails	Required	Agent capabilities and style
tools	list	[]	LangChain tools
memory_order	int	10	Messages to keep in memory
long_context	bool	True	Enable long-context mode
long_context_order	int	20	Context summarization threshold
persist_memory	bool	None	Enable persistent storage (requires memory_config in AgentManager)
plan	bool	False	Enable planner component
temperature	float	0.2	Sampling temperature

See docs/CONFIGURATION.md for all options.

---

Advanced Topics

Custom Tool Integration

from langchain.tools import tool

@tool
def calculate(expression: str) -> str:
    """Calculate mathematical expressions"""
    return str(eval(expression))

agent = manager.create_agent(
    agent_name="calculator",
    tools=[calculate]
)

Multi-Agent Orchestration

from masai.MultiAgents.MultiAgent import MultiAgentSystem, SupervisorConfig

# Decentralized MAS (peer-to-peer)
mas = MultiAgentSystem(agentManager=manager)
result = await mas.initiate_decentralized_mas(
    query="Complex task",
    set_entry_agent=agent1,
    memory_order=3
)

# Hierarchical MAS (supervisor-based)
supervisor_config = SupervisorConfig(
    model_name="gpt-4o",
    temperature=0.7,
    model_category="openai",
    memory_order=20,
    memory=True,
    extra_context={}
)

mas_hierarchical = MultiAgentSystem(
    agentManager=manager,
    supervisor_config=supervisor_config
)
result = await mas_hierarchical.initiate_hierarchical_mas(query="Complex task")

See docs/ADVANCED.md for advanced patterns.

---

API Reference

Core Classes

• AgentManager: Orchestrates agent creation and management
• Agent: Router-Evaluator-Reflector architecture
• MASGenerativeModel: LLM with memory management
• LongTermMemory: Persistent memory interface
• RedisConfig/QdrantConfig: Backend configuration

See docs/API_REFERENCE.md for complete API.

---

Troubleshooting

Redis Connection Refused

redis-server
# or
docker run -d -p 6379:6379 redis:latest

OpenAI API Key Not Found

export OPENAI_API_KEY="your-key-here"

Memory Not Being Retrieved

# Verify context overflow (access through LLM component)
print(f"Summaries: {len(agent.llm_router.context_summaries)}")
print(f"Long context order: {agent.llm_router.long_context_order}")
print(f"Persist memory: {agent.llm_router.persist_memory}")

See docs/TROUBLESHOOTING.md for more solutions.

---

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.

Support

• 📖 Documentation
• 🐛 Issues
• 💬 Discussions

---

Last Updated: October 31, 2025