AGI.green Framework

A modern Python framework for building AI-powered asynchonous chat applications with a browser interface, featuring markdown content rendering and a unified messaging system.

Application can be built entirely in Python or can integrate with arbitrary external services. This differs from conventional full stack solutions with a javascript frontend application making API calls to a Python backend. Instead, the frontend is a slave and the backend is the application.

Features

• Unified Message Protocol: Consistent message handling across different communication channels
• Rich Content Support:
  • Markdown rendering
  • Mermaid diagrams
  • MathJax equations
  • Code syntax highlighting
  • YAML forms implementing Vueform
• Flexible Architecture:
  • WebSocket-based real-time communication
  • Message Queue (AMQP) integration for distributed systems - rabbitmq|ABS|inprocess
  • Extensible protocol system - asynchronous interaction with anything
• AI Integration:
  • OpenAI API support
  • Anthropic Claude support
  • Custom model integration capability
• Asynchronous Execution:
  • Non limited to call and response
  • Can add chat messages in response to arbitrary events, e.g. incoming email or SMS or task completion
  • Can execute arbitrary code in response to a message or other events
  • No javascript required on the frontend!
• Vue Forms in Markdown (chat and docs):
  • Support for YAML and JSON forms in markdown
  • Support for forms in chat messages
  • Support for forms in documents
  • Asynchronous form data events with no javascript required on the frontend (python only)
  • That means no need to write any javascript to use forms!

Quick Start

1. Install dependencies:

# Install RabbitMQ
sudo apt-get install rabbitmq-server  # Ubuntu/Debian
brew install rabbitmq                 # macOS

# Install package
pip install agi.green

2. Create a basic chat application:

from agi_green.dispatcher import Dispatcher

class ChatNode(Dispatcher):
    async def on_mq_chat(self, author: str, content: str):
        'Receive chat message from RabbitMQ'
        await self.send_ws('append_chat', content=content)

    async def on_ws_chat_input(self, content: str = ''):
        'Handle chat input from browser'
        await self.send_mq('chat',
            author=self.name,
            content=content
        )

Protocol System

The framework uses a protocol-based architecture where each protocol handles specific types of communication:

• ws: WebSocket communication with browser
• mq: RabbitMQ for peer-to-peer messaging
• gpt: OpenAI API integration
• http: HTTP/HTTPS server
• Custom protocols can be added by extending the base Protocol class

Message Formatting

Support for rich content in messages:

# Markdown with syntax highlighting
await self.send_ws('append_chat',
    content='```python\nprint("Hello")\n```'
)

# Mermaid diagrams
await self.send_ws('append_chat',
    content='```mermaid\ngraph TD;\nA-->B;\n```'
)

# Math equations
await self.send_ws('append_chat',
    content=r'\\( \int x dx = \frac{x^2}{2} + C \\)'
)

Development Status

Currently in active development (March 2024). The framework is being used in production but the API may have breaking changes. See CHANGELOG for version history.

Requirements

• Python 3.11+
• RabbitMQ server
• Modern web browser

Contributing

Contributions are welcome! Please check our Contributing Guidelines for details.

License

Copyright (c) 2024 Ken Seehart, AGI Green. See LICENSE for details.