| **Building for a world with capable AI** | |
| While power over AGI and resources consolidates into the hands of a few, we're building for a different future. One where billions can govern powerful multi-agent systems with access to capital, physical resources, and real-world agency. | |
| ## Privacy-preserving personal AI | |
| Our current focus is on privacy-first personal AI that enable users to leverage powerful AI while keeping personal data truly private. Our models perform on-device anonymization, replacing personally identifiable information (PII) with semantically similar alternatives before queries ever leave your device. | |
| ## Our models | |
| **Anonymizer series** - Small language models (0.6B, 1.7B, 4B) trained for surgical PII replacement: | |
| - Detect and replace names, companies, financial data, addresses, and other sensitive information | |
| - Maintain query context and semantic meaning | |
| - Available in multiple formats: PyTorch, GGUF, MLX (iOS-optimized) | |
| - Trained using reinforcement learning with GPT-4.1 as judge | |
| ## Applications | |
| Our models power [Silo](http://link.freysa.ai/appstore), an iOS app that provides ChatGPT/Claude-quality responses while keeping your sensitive, personal information on-device. | |
| --- | |
| **Learn more:** [eternis.ai](https://www.eternis.ai/) | |
| **Contact:** contact@eternis.ai |