MiniAGI is a Python-based autonomous AI agent designed for general-purpose tasks, leveraging OpenAI's API (GPT-3.5-Turbo and GPT-4). It aims to execute complex objectives by breaking them down into steps, utilizing tools, maintaining short-term memory, and incorporating self-criticism for improved accuracy. The agent is suitable for users who want to automate tasks ranging from coding and content creation to system analysis and strategic planning.

How It Works

MiniAGI operates by receiving a high-level objective and then generating a sequence of actions. It employs a chain-of-thought process, internal monologue, and self-criticism to refine its plans. The agent can execute code (Python, shell), perform web searches, and interact with external APIs. Its memory is managed through summarization, allowing it to retain context over longer task sequences.

Quick Start & Requirements

• Install: git clone https://github.com/muellerberndt/mini-agi && cd mini-agi && pip install -r requirements.txt && cp .env_example .env
• Requirements: Python 3.10 or higher, OpenAI API key.
• Usage: python miniagi.py <objective>
• Docs: https://github.com/muellerberndt/mini-agi

Highlighted Details

• Supports GPT-3.5-Turbo and GPT-4 models.
• Features an optional "critic" module for enhanced accuracy at the cost of more API calls.
• Demonstrates capabilities in code generation, web scraping, system commands, and even simulated API interactions.
• Includes examples for complex tasks like penetration testing and world domination planning.

Maintenance & Community

The project appears to be maintained by muellerberndt. No specific community channels or roadmap details are provided in the README.

Licensing & Compatibility

The repository does not explicitly state a license in the provided README. Users should verify licensing for commercial use or integration into closed-source projects.

Limitations & Caveats

The agent may suggest harmful commands or code, requiring user caution. Data shared with third-party API providers like OpenAI is subject to their privacy policies. Some simulated actions (e.g., ordering pizza) do not involve real-world execution due to missing API keys or functionality.