AAF (Agentic AI Framework)

AAF is a versatile and extensible framework for building and managing agentic AI models. It provides a unified interface for various language model providers and implements advanced virtual models for complex, agent-like conversational scenarios.

Note that AAF is primarily a personal learning project focused on exploring agentic AI and LLM use, including complex multi-step interactions. While it can be useful for actual use cases like autonomous chat agents and multi-stage task completion, please exercise caution when considering it for anything even remotely important.

Features

• Support for multiple LLM providers (OpenAI, Anthropic, Ollama, LiteLLM) to act as the foundation for AI agents
• Advanced conversation management with Threads and Sessions for maintaining agent state
• Virtual models for complex, multi-step agent behaviors:
  • TwoPhase: For agents that plan before acting
  • Multiphase: For agents that can break down and tackle complex tasks
  • Router: For meta-agents that can delegate to specialized sub-agents
• Tool integration for function calling capabilities
• Cost and token usage tracking

Installation

pip install aaf

Quick Start

from aaf.threads import Session

thread = Session().create_thread("gpt-4o", system="You are a helpful assistant.")
thread.add_message("user", "What is the capital of France?")

async with thread.run() as stream:
    async for chunk in stream.text_chunks():
        print(chunk.content, end="", flush=True)
    print()

print(thread.cost_and_usage().pretty())

Usage

LLM Providers

AAF supports multiple LLM providers. To use a specific provider, specify the model name when creating a thread:

thread = session.create_thread("gpt-4o")  # OpenAI
thread = session.create_thread("claude-3-5-sonnet-20240620")  # Anthropic
thread = session.create_thread("llama3.1:8b")  # Ollama

Virtual Models

AAF implements several virtual models for advanced use cases:

• TwoPhase: Generates a prompt and then uses it to create a response
• Multiphase: Multi-step process for complex questions, including drafting, feedback, and refinement
• Router: Selects the appropriate model based on the user's request

Using a virtual model is same as with standard models:

from aaf.virtual_models.two_phase import TwoPhaseModel
from aaf.threads import Session

thread = Session().create_thread(model="two-phase", runner=TwoPhaseModel())
thread.add_message("user", "What is the capital of France?")

async with thread.run() as stream:
    async for chunk in stream.text_chunks():
        print(chunk.content, end="", flush=True)
    print()

print(thread.cost_and_usage().pretty())

Project Structure

• aaf/: Main package directory
  • llms/: LLM provider implementations
  • virtual_models/: Virtual model implementations
  • tools/: Tool definitions
  • threads.py: Thread and Session management
  • logging.py: Custom logging implementation
  • utils.py: Utility functions

License

This project is licensed under the MIT License. See the LICENSE file for details.