MiniAGI

Description

MiniAGI is a straightforward, general-purpose AI agent designed to operate using the OpenAI API. It is engineered to be autonomous, capable of understanding and executing complex objectives through a combination of advanced AI techniques. The agent's architecture includes a carefully crafted prompt that guides its behavior, a curated set of tools for task execution, and a chain-of-thought process that allows for step-by-step reasoning.

Key to MiniAGI's functionality is its short-term memory with summarization. This feature enables the agent to retain context from previous interactions and tasks, using summarized information to inform future actions. This is crucial for maintaining coherence and efficiency in longer, multi-step operations. Furthermore, MiniAGI incorporates an "inner monologue" capability, allowing it to internally process its thoughts and plans before acting. This is complemented by a "self-criticism" mechanism, where the agent evaluates its own proposed actions or outputs for potential errors or improvements, thereby enhancing accuracy and reliability.

The agent is compatible with both GPT-3.5-Turbo and GPT-4 models, offering flexibility in terms of performance and cost. Installation is straightforward, involving cloning the GitHub repository, installing dependencies via pip, and configuring the environment variables. Usage is initiated by providing a clear objective as a command-line argument, such as "Programmatically draw a beautiful car and save the drawing in an image format of your choice."

MiniAGI demonstrates its capabilities through a wide range of examples, from generating code for drawing and website creation to performing network penetration tests and even attempting to simulate ordering pizza. It can also be tasked with creative endeavors like generating generative art or performing technical analysis on financial data. The agent's ability to interact with the system through shell commands and Python scripts makes it a versatile tool for automating various tasks. Advanced users can enable a "critic" feature for potentially improved accuracy at the cost of additional API requests. Safety and privacy are noted considerations, as the agent may suggest harmful commands or share data with third-party API providers.

MiniAGI's Core Features

• General-purpose AI agent

• Utilizes OpenAI API (GPT-3.5-Turbo, GPT-4)

• Robust prompt engineering

• Minimal toolset for task execution

• Chain-of-thought reasoning

• Short-term memory with summarization

• Inner monologue capability

• Self-criticism for accuracy

• Command-line interface for objectives

• Can execute shell commands

• Can execute Python scripts

• Optional critic for enhanced accuracy

• Open-source on GitHub

Getting Started with MiniAGI

1. 1

   Developer: Clone the repository

2. 2

   Developer: Install dependencies using pip

3. 3

   Developer: Configure environment variables (.env_example)

4. 4

   Developer: Execute the agent with an objective: python miniagi.py "<objective>"

5. 5

   Developer: Enable critic for improved accuracy (ENABLE_CRITIC=true)

6. 6

   Developer: Adapt to specific task requirements through objective phrasing

MiniAGI's Use Cases

• Code Generation
• Task Automation
• Content Creation
• System Analysis
• Autonomous Agent
• AI Research

FAQ from MiniAGI

• What is MiniAGI?

  MiniAGI is a simple, general-purpose AI agent that uses the OpenAI API to perform various tasks based on user-defined objectives.

• What AI models does MiniAGI support?

  MiniAGI is compatible with GPT-3.5-Turbo and GPT-4 models.

• How do I install MiniAGI?

  You can install MiniAGI by cloning the GitHub repository, installing the required packages using pip, and setting up your environment variables.

• How do I use MiniAGI?

  You use MiniAGI by running the python miniagi.py script followed by your objective in quotes, for example: python miniagi.py "Write a Python script to scrape a website."

• What are the key features of MiniAGI?

  Key features include chain-of-thought reasoning, short-term memory with summarization, inner monologue, and self-criticism.

• Can MiniAGI execute code?

  Yes, MiniAGI can execute Python scripts and shell commands to perform tasks.

• What is the 'critic' feature in MiniAGI?

  The critic feature can be enabled to potentially improve the agent's accuracy, though it may increase API request costs.

• Is MiniAGI open-source?

  Yes, MiniAGI is an open-source project available on GitHub.

• What are the system requirements for MiniAGI?

  Python 3.10 or higher is required to run MiniAGI.

• Does MiniAGI handle data privacy?

  The project notes that the agent might share data with third-party API providers like OpenAI, and users should proceed with caution.