Show HN: A Manifesto for a Privacy-First, Open Core AI Wearable (GitHub) (github.com)

A solo founder has posted a "Manifesto" and prototype concept for "Companion," a privacy-first, always-on AI wearable designed to act as a personal assistant or coach while keeping raw personal data on-device. The core claim: all audio and video understanding runs 100% locally — the device is "deaf and blind" by default and only activates narrowly scoped, user-assigned "Purpose-Driven Filters" (e.g., nutrition tracking) when explicitly tasked. To prove the privacy promise, the team plans an open-source core (OS and data pipeline) so third parties can verify that nothing leaves the device without consent. The startup is in concept phase, soliciting feedback and early interest ahead of a YC application. This approach is significant because it shifts always-on agents away from cloud-first architectures that erode trust; it foregrounds verifiability and a business model that eschews data monetization in favor of hardware/software sales or optional cloud features. Technically, it implies heavy investment in on-device ML — efficient audio/vision models, low-power hardware, model compression, and secure update mechanisms — and raises trade-offs between local latency/privacy and cloud capabilities. Open-sourcing the core increases auditability but also exposes implementation details and demands robust engineering to meet real-world constraints (battery, compute, personalization). If executed, the project could advance privacy-preserving, verifiable agent designs and shape expectations for trust in ambient AI.