Show HN: Symbio self fine-tuning AI loop (github.com)

Symbio, a new local AI tool, introduces a self fine-tuning mechanism that learns from user corrections with no reliance on cloud services or subscriptions. This innovative AI allows users to train it on-the-fly by simply correcting its mistakes during interactions. As corrections accumulate, they’re documented as markdown notes. Once a defined threshold is reached, Symbio uses this data to fine-tune a specialized model adapter for improved accuracy in future tasks. The significance of Symbio lies in its Mixture of Agents (MOA) architecture, which distributes tasks among smaller worker models, each learning from errors with guidance from a headmaster model. This approach minimizes repetitive training while maximizing personalization. Additionally, the AI runs locally on Apple Silicon using the MLX framework, ensuring privacy and reduced costs typical of cloud-based AI solutions. The platform supports various functionalities, including executing shell commands and managing notes, demonstrating a self-sufficient, user-friendly AI experience.