Rewisp – An ambient screen memory for macOS that never writes images to disk (yashmitb.github.io)

Rewisp, a new ambient memory application for macOS, has been launched with a novel approach to managing user data while prioritizing privacy. Unlike traditional memory tools that save images and data to the cloud, Rewisp operates locally without retaining screenshots. Instead, it utilizes Apple’s Vision framework to analyze screen content in real-time, converting it to text while discarding pixel data before saving any information. This means users can receive responses on-the-spot by hitting ⌘⇧Space, allowing them to interact with the app through natural language and manage their information directly and securely. The significance of Rewisp lies in its commitment to user privacy and its innovative use of AI models. It starts with a free on-device model and escalates to more powerful models like Gemini or Claude only when necessary, ensuring that queries are answered efficiently without compromising user data. Additionally, it offers features such as a digest of the day's activities, smart triggers to capture critical moments, and a Vault for storing sensitive information like resumes, all secured behind user authentication. This shift towards privacy-centric, on-device computation presents a compelling advancement for the AI/ML community, particularly in enhancing user trust in AI applications.