BLUEPRINT

Design Principles

  1. Out-of-the-Box — Regular users can get started with three commands after pip install. No machine learning knowledge or GPU driver configuration required.
  2. Fully Automated — From data collection to model output, no manual intervention needed. Like a washing machine: put in clothes, press start, take out clean clothes.
  3. Multi-Version Adaptation — One training run produces multiple versions for different computing capabilities. The same "you" can simultaneously live on a smartwatch, phone, and computer.
  4. Repeatable Execution — Retrain after data updates, and Ego evolves accordingly. As you grow, Ego grows too.
  5. Swappable Base — Switch base models when new ones are released, retrain LoRA. Like upgrading a phone's processor — performance improves but "personality" stays the same.
  6. Plugin Extensible — Data source adapters, base models, and editions are all extensible. The community can contribute new data source plugins (e.g., Notion adapter, WhatsApp chat history importer).
  7. Privacy First — All training can be completed locally, data never leaves the device. Your diary, chat history, and work notes stay forever on your own computer.