cognira is a self-learning AI chat assistant served independently on a single iMac using MLX inference. Users can teach it preferences or facts mid-conversation, which are saved to private memory and applied from the next reply onward. The hosted model, cognira Retrain 12B, is built on a Gemma 4 base and is not forwarded to any external API. A local version called cognira Entity runs the same models entirely on the user's own hardware with optional weight-level retraining. An OpenAI-compatible API endpoint is also available for developers at a token-based price.