MultiAgents Framework

🚀 LLM-Powered Multi-Agent Orchestration Framework

A hybrid event-driven orchestration framework designed specifically for LLM agents and AI developers. Build intelligent, conversational, and autonomous multi-agent systems with built-in DSPy integration, real-time collaboration, and production-ready tooling.

⚠️ Experimental: This framework is in active development. APIs may change between versions.

🎯 Built for LLM Agents

Perfect for AI developers building:

• 🤖 Conversational AI Systems - Multi-agent chat with intelligent routing
• 🔍 Research Assistants - Collaborative research with specialized agents
• 🧠 LLM-Driven Workflows - Dynamic orchestration with real-time decisions
• 🛠️ Tool-Using Agents - Multi-modal agents with external integrations
• 📊 Data Analysis Pipelines - Intelligent data processing with LLM insights

✨ Key Features

• 🧠 LLM-First Design: Built-in DSPy integration with Gemini, GPT, Claude support
• 💬 Conversational AI: Intelligent routing between conversation and task execution
• 🔧 Tool Integration: Easy integration with web search, calculators, APIs
• 🎭 Multi-Agent Collaboration: Specialized agents working together seamlessly
• 📊 Production Monitoring: Complete observability designed for LLM workflows
• 🔄 Event-Driven: Scalable async communication perfect for AI workloads
• 🛡️ Fault Tolerance: Built-in compensation and rollback for complex workflows

Quick Start

Prerequisites

1. Python 3.8+
2. Redis Server (for event bus and state storage)

Start Redis:

# macOS with Homebrew
brew services start redis

# Ubuntu/Debian
sudo systemctl start redis

# Docker
docker run -d -p 6379:6379 redis:alpine

Installation

Install from PyPI (Recommended):

# Install the framework
pip install multiagents-framework

# Or with uv (faster)
uv add multiagents-framework

Development Installation:

1. Clone the repository:
git clone https://github.com/xavierau/multiagents.git
cd multiagents

2. Create virtual environment:
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install in development mode:
pip install -e .

Verify Installation:

multiagents --version
multiagents --help

🚀 Quick Start Examples

Run Interactive Examples:

# Interactive example menu
python run_examples.py

# Or run specific examples:
multiagents --examples

Available Examples:

1. 💬 Smart Research Assistant - LLM-powered conversational research with web search
2. 🤖 Interactive Chatbot - Multi-personality conversational AI with DSPy
3. 🛒 E-commerce Workflow - Complete order processing with compensations
4. 📊 Monitoring Demo - Production-ready observability features

Try the Smart Research Assistant:

# Navigate to example (if installed from source)
cd examples/smart_research_assistant
python cli.py

# Ask questions like:
# "What are the latest trends in renewable energy?"
# "Calculate compound interest on $5000 at 8% for 3 years"
# "Hi!" (conversational mode)

Framework Overview

Core Components

1. Orchestrator: Manages workflow state and coordinates activities
2. Workers: Stateless task executors created with simple decorators
3. Event Bus: Decoupled communication layer (Redis Pub/Sub)
4. Monitoring: Comprehensive observability and debugging

Basic Usage

from multiagents import (
    Orchestrator, WorkflowBuilder, WorkerManager, worker
)
from multiagents.event_bus.redis_bus import RedisEventBus

# Define workers
@worker("process_data")
async def process_data_worker(context):
    data = context["input_data"]
    # Process the data
    return {"processed": data, "timestamp": "2024-01-01T00:00:00Z"}

# Create workflow
workflow = (WorkflowBuilder("data_processing")
    .add_step("process", "process_data")
    .build())

# Set up framework
event_bus = RedisEventBus()
worker_manager = WorkerManager(event_bus)
orchestrator = Orchestrator(workflow, event_bus)

# Register worker and start
worker_manager.register(process_data_worker)
await event_bus.start()
await worker_manager.start()
await orchestrator.start()

# Execute workflow
transaction_id = await orchestrator.execute_workflow(
    "data_processing",
    {"input_data": "example data"}
)

Monitoring & Observability

The framework includes comprehensive monitoring:

from multiagents.monitoring import MonitoringConfig, EventMonitor, WorkerMonitor

# Setup monitoring
config = MonitoringConfig()
logger = config.create_logger()
event_monitor = EventMonitor(logger=logger)
worker_monitor = WorkerMonitor(logger=logger)

# Integrate with framework
event_bus = RedisEventBus(event_monitor=event_monitor, logger=logger)
worker_manager = WorkerManager(event_bus, worker_monitor=worker_monitor, logger=logger)

Monitoring Features:

• Event Lifecycle Tracking: Complete event journey from dispatch to completion
• Worker Performance: Success rates, processing times, health monitoring
• Structured Logging: JSON logs with automatic rotation
• Error Tracking: Detailed error context and failure patterns
• Real-time Metrics: Performance dashboards and alerting

Configuration (monitoring.yaml):

logging:
  default_logger: "file"
  level: "INFO"
  file_path: "./logs/multiagents.log"

event_monitoring:
  enabled: true
  trace_retention_hours: 24

worker_monitoring:
  enabled: true
  health_check_interval_seconds: 30

📋 Example Applications

1. 💬 Smart Research Assistant

LLM-powered conversational research system

• Intelligent routing between conversation and research
• Multi-agent collaboration (Coordinator, Researcher, Analyst, Formatter)
• Real Google Custom Search API integration
• Gemini LLM with DSPy optimization
• Production-ready configuration system

2. 🤖 Interactive Chatbot

Multi-personality conversational AI

• DSPy-powered natural conversations
• Configurable personalities and responses
• Context-aware conversation management
• Real-time interaction with memory

3. 🛒 E-commerce Order Processing

Production-ready order workflow

• Multi-step order validation and processing
• Intelligent inventory and payment handling
• Automatic compensation and rollback
• LLM-generated confirmations and notifications

4. 📊 Production Monitoring

Enterprise-grade observability

• Real-time agent performance monitoring
• Event lifecycle tracking and debugging
• Structured logging with automatic rotation
• Error pattern analysis and alerting

Development

Project Structure

multiagents/
├── orchestrator/          # Workflow orchestration
├── worker_sdk/           # Worker development SDK
├── event_bus/            # Event bus implementations
├── monitoring/           # Observability system
├── core/                 # Core utilities
└── examples/             # Example implementations

Key Design Principles

• SOLID Principles: Clean, maintainable architecture
• Event-Driven: Fully decoupled communication
• Fault Tolerance: Built-in compensation and rollback
• Observability: Comprehensive monitoring and debugging
• Developer Experience: Simple, intuitive APIs

Common Commands

# Install dependencies
pip install -r requirements.txt

# Run examples
python run_examples.py

# Run tests (when available)
pytest

# Format code
black .
ruff check --fix .

# Type checking
mypy multiagents/

Architecture

The framework follows a hybrid orchestration/choreography pattern:

• Orchestration: Centralized workflow management with state machine
• Choreography: Decoupled event-driven communication between components
• Saga Pattern: Distributed transaction management with compensations
• Event Sourcing: Complete audit trail of all activities

This approach provides the benefits of both patterns:

• Centralized Logic: Easy to understand and debug workflows
• Decoupled Components: Scalable and resilient architecture
• Fault Tolerance: Automatic compensation and recovery
• Observability: Complete visibility into system behavior

Contributing

1. Follow SOLID principles and clean code practices
2. Add comprehensive monitoring to all components
3. Include tests for new functionality
4. Update documentation and examples

🔗 LLM Agent Integration

Claude Code Integration: This framework includes specialized Claude Code subagents for enhanced development:

# Auto-install Claude subagent (when using Claude Code)
multiagents install-agent claude-code

# Or manually copy the agent
cp multiagents/agents/multiagents.md ~/.claude/agents/

llms.txt Support: This project includes llms.txt for LLM consumption with documentation pointers to GitHub for always up-to-date information.

🌐 Community & Support

• GitHub: https://github.com/xavierau/multiagents
• PyPI: https://pypi.org/project/multiagents-framework/
• Issues: Report bugs and request features
• Discussions: Community discussions and help

📜 License

MIT License - see LICENSE file for details.

---

Built with ❤️ for the LLM agent community
Making multi-agent AI systems accessible, reliable, and production-ready.