Rewisp — an ambient memory for your Mac: See it once. Ask forever.
Rewisp is a new Mac application designed to act as an "ambient memory," passively reading screen text on-device to remember information, track changes, and manage commitments, all while prioritizing user privacy.
Intelligence analysis by Gemini 2.5 Flash
The Rewisp app aims to alleviate the burden of information overload by continuously capturing and indexing text displayed on a user's Mac screen. It enables semantic search, tracks numerical data over time, and provides reminders for commitments, operating with a strong emphasis on local processing and data privacy.
Imagine your Mac has a super-smart brain that quietly reads everything on your screen, like a helpful friend. It remembers important stuff, reminds you of promises you saw, and helps you find things you saw before, all without sending your secrets anywhere. It's like having a perfect memory for your computer.
Analysis
Enhancing Personal Productivity with Ambient AI
Rewisp introduces an innovative approach to personal productivity by transforming a Mac into an "ambient memory." The application continuously reads text displayed on the user's screen, indexing it for future retrieval and analysis. This passive capture mechanism is designed to address the common problem of losing track of small but important details scattered across various applications, messages, and web pages. Key features include the ability to "catch promises" like "I'll send it Friday" and provide timely reminders, as well as tracking changes on a page over time, allowing users to ask "what changed since Tuesday?" The app also supports semantic search, enabling users to find information based on meaning rather than just keywords, and can track numerical data like weight or grades.
Privacy-First Design and Local AI Processing
A core tenet of Rewisp's design is its commitment to user privacy. The application processes all screen text on-device, ensuring that sensitive information never leaves the user's Mac by default. Screenshots are discarded immediately after text extraction, and all data is stored locally in a single SQLite file. The developer, Yashmit Bhaverisetti, emphasizes that nothing leaves the device unless explicitly requested by the user. For its AI reasoning layer, Rewisp primarily utilizes Apple's on-device models. While it offers the option to connect to external AI providers like Claude, ChatGPT, or Gemini for more complex queries, even then, only the question and relevant text snippets are sent, never full screenshots or complete history. The app also includes privacy controls such as pausing capture, a "nuke" option for recent data, and hard-coded exclusions for sensitive applications like banking or messaging apps.
Technical Underpinnings and Scalability
From a technical standpoint, Rewisp is engineered for efficiency and scalability, despite its local-first approach. The database, stored in an SQLite file, remains compact due to text-only storage and deduplication of near-identical frames, growing by only a few MB per day. This design ensures that months of usage result in tens to low-hundreds of MB, not gigabytes. Keyword search leverages FTS5 and bm25 (an inverted index) for sub-millisecond query times, regardless of database size. Semantic search, while currently using brute-force cosine similarity over embeddings, remains fast for the application's retention cap of approximately 20-35k rows. The system also incorporates built-in pruning with a six-month retention policy that is importance-based, not just clock-based, and nightly consolidation processes that fold old sessions into compact summaries. This intelligent data management ensures the database becomes leaner and higher-signal over time, preventing unbounded growth.
Key points
- Rewisp is a Mac app that functions as an "ambient memory" by reading and remembering screen text on-device.
- It offers features like promise-catching, tracking changes on pages over time, and semantic search capabilities.
- The app prioritizes privacy, processing all data locally, discarding screenshots instantly, and storing information in a private SQLite file.
- It is free and open-source (MIT license), utilizing Apple's on-device AI models by default, with optional integration for external AI engines.
- Data storage is optimized for efficiency, with text-only storage, deduplication, and a six-month retention policy that consolidates old data into summaries.
Rewisp could significantly enhance personal productivity by automating the recall of small but crucial details, ensuring commitments are met, and making information retrieval effortless. Its strong privacy focus, combined with local AI processing, could build user trust and encourage wider adoption for sensitive personal data management.
Despite robust privacy measures, some users may still be hesitant about an application constantly monitoring their screen, potentially limiting its reach. As a solo-developed, early-stage project, long-term maintenance, feature expansion, and competitive pressures could pose challenges for its sustained development and support.
