conw.ai

Conw is an AI chatbot that learns directly from the people using it. Instead of relying on a static model that never improves mid-conversation, Conw lets you correct its replies, rate responses, or explain a new word — and that knowledge sticks immediately. The tool runs its own 188M-parameter model called Conway-Omega, served locally on a single 16GB iMac using Apple's MLX framework. It is not a wrapper around someone else's API; the inference and learning happen on Conw's own hardware.

What makes Conw stand out is its transparent learning loop. When you teach it something, the correction enters your private memory right away and can be used in the same chat. But before any change affects the shared model, it goes through a guarded queue: only safe, useful, short replies become candidates for a background update. Training only starts when the server is idle, so live chat always gets priority. The site shows you whether a lesson passed or failed, and unsafe or low-rated replies never make it into the model.

Conw is best for curious users who want to shape an AI's behavior through direct feedback rather than prompt engineering. It is also interesting for developers and AI enthusiasts who appreciate a small, independently-run system that prioritizes transparency over marketing hype. Use cases include learning a new language by teaching vocabulary, refining an AI assistant's tone for personal projects, or simply exploring how a continuously learning model behaves in practice.