AGI-3

Artificial General Intelligence

Overview

This repository contains an implementation of a modular-agentic architecture for Artificial General Intelligence (AGI). The architecture is designed to facilitate the development of autonomous agents capable of complex reasoning, learning, and interaction with their environment.

Example

from agi.sophos import Agent
from agi.sophos_tools import *

toolbox = all_tools()

agent = Agent(
    name="Sophos Agent",
    instructions="You are an AI Agent.",
    model="agi/Qwen3-4B-Instruct-2507-Q3_K_S.gguf",
    tools=toolbox,
)

prompt = "Roll a dice, also whats the weather in Tokyo?"
response = agent.run(prompt)
print(response) 

Output:

"""
You rolled a 4 on a 6-sided die. The weather in Tokyo is sunny with a temperature of 12°C during fall.
"""

Research Paper

Read the research paper

This paper delineates a comprehensive architectural framework for the progressive realization of Artificial General Intelligence (AGI), predicated upon a modular-agentic paradigm. We present a system design that integrates sophisticated tool-use capabilities, hierarchical memory management, dynamic code execution, and nascent world-modeling functionalities. The proposed architecture, exemplified through a lightweight Qwen3-4B-Instruct-2507-Q3_K_S.gguf model, demonstrates a robust foundation for emergent cognitive properties such as autonomy, recursive self-improvement, and goal-oriented behavior. Furthermore, we explore the theoretical underpinnings of consciousness as an emergent property within complex neural architectures and postulate pathways towards super-intelligence through advanced computational and embodied interaction modalities. The exposition maintains a rigorous academic tone, employing advanced terminology to articulate the intricate conceptual and technical facets of AGI development.

Implementation

This is a more advanced implementation of the original AGI repository. It includes more tools, better memory management, and a more advanced agent structure.