Phoenix Agents

A framework for building AI agents with modular connectors and LLM integration.

Installation

pip install phoenix-agents

Features

• Modular connector system for agent interactions
• Built-in support for Azure AI and OpenAI
• Chat history management
• Extensible agent architecture
• Session management for multi-user scenarios

Quick Start

Here's a simple example using Phoenix to create a simple AI agent:

from dotenv import load_dotenv
import asyncio
import os

from phoenix.agent import Agent
from phoenix.user_session import UserSession
from phoenix.models.azure_ai_inference import AzureAIInferece
import phoenix.models.openai_history as openai_history

load_dotenv()

async def main():
    # Initialize chat history
    chat_history = openai_history.ChatHistory()

    # Setup LLM with Azure AI
    llm = AzureAIInferece(
        token=os.getenv("GITHUB_TOKEN"),
        history=chat_history
    )

    # Create agent
    agent = Agent(
        brain=llm,
        history=chat_history,
    )

    # Create user session and interact
    session = UserSession()
    response = await agent.call("Hello, how are you?", session)
    print(response)

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

Here's a simple example using Phoenix to create a moody AI agent with MCP connectors (The full example can be found in Moody AI Repository):

from dotenv import load_dotenv
import asyncio
import os

from phoenix.agent import Agent
from phoenix.user_session import UserSession
from phoenix.models.azure_ai_inference import AzureAIInferece
import phoenix.models.openai_history as openai_history
from phoenix.connectors.mcp import MCPClient, MCPServer

load_dotenv()

async def main():
    # Initialize chat history
    chat_history = openai_history.ChatHistory()

    # Setup LLM with Azure AI
    llm = AzureAIInferece(
        token=os.getenv("GITHUB_TOKEN"),
        history=chat_history
    )

    # Setup MCP connector with servers
    mcp = MCPClient([
        MCPServer(path="path/to/mood.py")
    ])

    try:
        await mcp.connect()

        # Create agent
        agent = Agent(
            brain=llm,
            history=chat_history,
            connector=mcp,
            system="You are a moody AI, you need to know your current mood to know how to respond.",
        )

        # Create user session and interact
        session = UserSession()
        response = await agent.call("Hello, how are you?", session)
        print(response)

    finally:
        await mcp.cleanup()

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

Configuration

The framework uses environment variables for configuration. Create a .env file with:

GITHUB_TOKEN=your_token_here

Contributing

Contributions are welcome!

License

MIT

Documentation

TODO