This project provides a GPT-4 client that enhances the AI's capabilities by giving it long-term memory and the ability to write and execute its own plugins. It targets users who want to leverage GPT-4 for more complex, self-improving tasks, enabling it to recall information beyond its context window and overcome limitations like poor mathematical abilities through self-generated tools.

How It Works

GPTChat implements a long-term memory system, allowing GPT-4 to store and retrieve information across conversations, overcoming the limitations of its fixed context window. A key feature is its plugin architecture, which enables GPT-4 to autonomously write, compile, and execute code (plugins) to perform tasks it's not inherently good at, such as complex calculations or random number generation. This self-improvement loop allows GPT-4 to combine memory recall and plugin execution for multi-step task completion.

Quick Start & Requirements

• Install Go from https://go.dev/.
• Obtain an OpenAI API key with GPT-4 access from https://platform.openai.com/account/api-keys.
• Set the OPENAI_API_KEY environment variable.
• Run with go run . from the gptchat directory.
• Plugin functionality requires Unix-based systems (Linux, macOS) or WSL2 on Windows.

Highlighted Details

• GPT-4 can write its own plugins for self-improvement.
• Enables multi-step task completion by combining memory and plugins.
• Features a long-term memory system for information recall beyond context windows.
• Runs in a supervised mode by default, requiring user confirmation before executing GPT-generated code.

Maintenance & Community

• Contributions via Pull Requests are welcome.
• Users are cautioned about prompt changes potentially disrupting GPT-4's command usage.

Licensing & Compatibility

• License: See LICENSE.md (likely permissive, but requires verification).
• Compatibility: Plugin execution is restricted to Unix-like environments or WSL2.

Limitations & Caveats

GPT-generated code is untrusted and potentially dangerous, executing with the user's permissions; running in a container or VM is recommended. Disabling supervised mode is strongly discouraged.