GrimmBot

GrimmBot is an advanced AI agent designed to learn and improve over time. It operates within a safe, isolated environment called a Debian Docker container, meaning it won't interfere with your main computer system. This AI can run commands on your computer and browse the internet. A key feature is its ability to learn from its mistakes. If it makes an error, it creates new rules to avoid repeating that mistake in the future. It can also create new tools for itself if the existing ones aren't enough for a task.

GrimmBot remembers past interactions and tasks using a system called RAG memory. It can also schedule tasks and improve its performance based on your feedback through a process called Unsloth 2.0 LoRA training. For sensitive actions, it has a permission system. You can even see what the AI is doing on your desktop in real time using noVNC.

All the information GrimmBot learns, like its custom tools and improvements, is stored in portable folders. This makes it easy to move its learned knowledge to another computer. Files are shared between your computer and the AI's environment through a special folder calledwormhole.

To set up GrimmBot, you need Docker Desktop. After downloading the project, you'll need to set up a configuration file with details about the AI model you want to use. If you plan to train the AI with your feedback using local models, you'll need a specific Python version installed. Once set up, you start GrimmBot using a command, and then access it through your web browser.

To help GrimmBot improve, you can give it feedback by rating its responses. After you provide enough feedback, you can click an 'Adapt' button. The system will then fine tune the AI with your input. After restarting the AI, it will use these new improvements.

GrimmBot can work with many different AI models, including those that run offline. Some features, like understanding images, require AI models that support them. The AI's self-improvement works by looking at errors, writing rules to prevent them, and trying again, creating a cycle of learning.

While the project doesn't accept direct code changes through pull requests, users are encouraged to report bugs or suggest new features through GitHub issues. GrimmBot is available under the GNU AGPL V3 License.