🤖 MiniAgent Framework

A simple, powerful agent framework for building conversational AI agents with tool use, streaming, Redis-backed sessions, and robust retry logic.

📁 Project Structure

miniagent_framework/
├── core/                    # Core agent framework
│   ├── __init__.py         # Package exports
│   ├── core.py             # Agent, Thread, AgentConfig
│   ├── llm.py              # LLM client with retry logic
│   ├── tools.py            # Tool system and registry
│   └── events.py           # Event system and callbacks
│
├── extensions/             # Extended functionality
│   ├── redis_client.py     # Redis session manager
│   ├── session.py          # Session-aware agent
│   └── events_enhanced.py  # Enhanced events for sessions
│
└── tools/                  # Enhanced tool implementations
    └── tools_enhanced.py   # Token-optimized tools

demos/                      # Example usage and demos
├── demo_miniagent.py       # Basic framework demo
├── demo_redis_session.py   # Redis sessions demo
└── demo_enhanced_tools.py  # Token optimization demo

tests/                      # Test suites
└── test_comprehensive.py   # Comprehensive framework tests

docs/                       # Documentation
├── README_FRAMEWORK.md     # Core framework docs
├── README_REDIS.md         # Redis session docs
└── improvements.md         # Future improvements

config/                     # Configuration files
└── docker-compose.yml      # Redis setup

🚀 Quick Start

1. Install Dependencies

# Core framework only (choose your LLM provider)
pip install python-dotenv

# For OpenAI/GPT:
pip install openai

# For Google Gemini:
pip install google-generativeai

# For Anthropic Claude:
pip install anthropic

# Full installation with all providers:
pip install -r requirements.txt

2. Set Up Environment

# Create .env file with your provider's API key
echo "OPENAI_API_KEY=your-key-here" >> .env      # For OpenAI
echo "GOOGLE_API_KEY=your-key-here" >> .env       # For Gemini
echo "ANTHROPIC_API_KEY=your-key-here" >> .env    # For Claude
echo "TAVILY_API_KEY=your-key-here" >> .env       # Optional for web search

3. Run Demos

Basic Agent Demo

python demos/demo_miniagent.py

Redis Session Management

# Start Redis first
docker-compose -f config/docker-compose.yml up -d

# Run session demo
python demos/demo_redis_session.py

Token-Optimized Tools

python demos/demo_enhanced_tools.py

Multi-LLM Provider Demo

python demos/demo_multi_llm_providers.py

✨ Key Features

Core Framework

• 🤖 Multi-LLM Support - OpenAI, Google Gemini, Anthropic Claude
• 🔧 Tool System - Easy tool creation with decorators
• 📡 Real-time Streaming - Stream responses word-by-word
• 🔄 Retry Logic - Configurable retry strategies
• 📝 Event System - Complete audit trail of all actions

Extensions

• 💾 Redis Sessions - Persistent conversations across restarts
• 🚀 Token Optimization - Send only required params to LLM
• 🔐 Session Management - User-based session isolation
• ⚡ Caching - Cache tool results and LLM responses
• 🔄 Distributed - Multiple agent instances can share sessions

🤖 Multi-LLM Provider Support

The framework supports multiple LLM providers out of the box:

Supported Providers

Provider	Models	Environment Variable
OpenAI	GPT-4, GPT-3.5	OPENAI_API_KEY
Google Gemini	Gemini Pro, Gemini Flash	GOOGLE_API_KEY or GEMINI_API_KEY
Anthropic	Claude 3 (Opus, Sonnet, Haiku)	ANTHROPIC_API_KEY

Usage Example

from miniagent_framework.core import Agent, AgentConfig

# Using OpenAI (default)
agent = Agent(AgentConfig(
    provider="openai",
    model="gpt-4o-mini"  # Optional, uses default if not specified
))

# Using Google Gemini
agent = Agent(AgentConfig(
    provider="gemini",
    model="gemini-1.5-flash"
))

# Using Anthropic Claude
agent = Agent(AgentConfig(
    provider="anthropic",
    model="claude-3-haiku-20240307"
))

# The API is the same regardless of provider
response = await agent.run("Hello, how are you?")

Provider-Specific Features

All providers support:

• ✅ Streaming responses
• ✅ Tool/function calling
• ✅ Retry policies
• ✅ Temperature and max_tokens control

Switching Providers

You can easily switch providers without changing your code:

# Configure via environment variable
config = AgentConfig(
    provider=os.getenv("LLM_PROVIDER", "openai")  # Flexible provider selection
)

📚 Documentation

• Core Framework Documentation
• Redis Session Management
• Future Improvements

🧪 Testing

Run the comprehensive test suite:

python tests/test_comprehensive.py

🏗️ Architecture

The framework follows a modular architecture:

1. Core Layer - Basic agent functionality
2. Extensions Layer - Optional enhancements (Redis, caching)
3. Tools Layer - Reusable tool implementations
4. Application Layer - Your custom implementation

🤝 Contributing

1. Keep code simple and readable
2. Add tests for new features
3. Update documentation
4. Follow existing patterns

📄 License

MIT

---

Built with ❤️ for creators who want powerful AI agents without complexity.