Rewisp

Product Introduction

1. Definition: Rewisp is a native macOS menu bar application that functions as an ambient, on-device memory system. It is a privacy-first screen text capture and semantic search tool that operates locally on Apple Silicon Macs.
2. Core Value Proposition: Rewisp exists to solve digital amnesia by automatically and privately remembering all text displayed on your screen, enabling users to retrieve information through natural language queries instead of manual note-taking or frantic searching through browser history and documents.

Main Features

1. Ambient Text Capture & Commitment Tracking: Rewisp uses smart triggers (app switches, page loads, scroll pauses) to capture "wisps" – transient screenshots processed instantly in memory by Apple's Vision framework to extract text. It specifically parses this text for commitments (e.g., "I'll send it Friday") using natural language processing, creates reminders, and surfaces them on the due date.
2. Semantic Search & Contextual Recall: The application builds a local, vector-based search index of all captured text. This allows for typo-tolerant, meaning-based searches. Users can ask "that article about burnout" and Rewisp will find relevant content even if the exact keywords aren't present, a functionality impossible for simple text search or screen recording.
3. Privacy-First Architecture & Local AI Processing: Screenshots are never written to disk; text extraction happens in RAM. All data is stored in a local SQLite database. It employs a cascading AI chain: queries first attempt to be answered by Apple's on-device model or a user-installed local LLM (like Llama). Only if these fail is an optional, user-configured cloud model (Gemini, Claude) prompted, with only the query text sent.
4. Temporal Diffing & Change Detection: Rewisp versions the textual content of pages (e.g., assignment details, documentation). Users can ask "what changed since Tuesday?" and the system performs a diff operation, highlighting exact additions, deletions, and modifications, providing precise change tracking over time.
5. Secure Autofill (Vault) & Data Structuring: A Touch ID-secured "Vault" allows users to store personal details (name, address, standard responses). Rewisp can read form fields on web pages and auto-fill from the Vault, explicitly refusing to handle passwords or credit card numbers. It also structures unstructured data, like tracking numeric values (weight, prices) over time to generate sparklines.

Problems Solved

1. Pain Point: The "I saw it but can't find it" problem. Critical information slips away into the abyss of endless tabs, chat history, and documents, leading to wasted time and missed commitments.
2. Target Audience: Knowledge workers, students, researchers, project managers, and any Mac power user who handles high volumes of textual information across multiple applications (Slack, browsers, IDEs, documents) and needs reliable, private recall.
3. Use Cases: Recalling a specific statistic from a research paper viewed hours ago; tracking changes to a project brief over several days; automatically being reminded of a casual promise made in a chat; quickly filling out a repetitive web form with personal data; understanding how much time was actually spent in different apps during a workday.

Unique Advantages

1. Differentiation: Unlike cloud-based screen recorders or note-taking apps, Rewisp is architected for absolute privacy with on-device processing. Unlike manual bookmarking or note-taking, it requires zero upfront effort—memory is passive. Its semantic search is more powerful than macOS Spotlight, which is limited to file names and metadata.
2. Key Innovation: The "privacy by construction" model: decoupling memory from surveillance by discarding pixels instantly and keeping all data local. The integration of a cascading, cost-aware AI query chain that prioritizes free, local models before escalating to paid cloud APIs. The implementation of Model Context Protocol (MCP) server support, allowing other AI agents (like those in Claude Desktop or Cursor) to read from the user's local memory context.

Frequently Asked Questions (FAQ)

1. Is Rewisp a screen recorder that saves videos? No. Rewisp is not a screen recorder. It takes temporary, in-memory screenshots, uses Apple's Vision framework to perform Optical Character Recognition (OCR) to extract text, and immediately discards the image. Only the textual content is stored locally.
2. How does Rewisp's privacy compare to other "memory" apps? Rewisp is fundamentally different. Most competitors upload screenshots or recordings to the cloud for processing. Rewisp performs all processing on your Mac. Text is stored in a local SQLite database in your user folder. Cloud AI use (like Gemini) is optional and sends only the specific query text, not your memory history.
3. Can Rewisp capture passwords or private information? Rewisp is designed to avoid this. It has a configurable "kill list" to pause capture entirely for sensitive apps (e.g., password managers, banking sites). Furthermore, it actively strips patterns like credit card numbers and Social Security Numbers from captured text before storage.
4. What are the system requirements for Rewisp? Rewisp requires macOS 13.0 (Ventura) or later and is optimized for Apple Silicon Macs (M1/M2/M3). The use of Apple's on-device Vision framework and AI models necessitates these modern hardware and OS requirements.
5. How does the AI query chain work? When you ask a question, Rewisp first attempts to answer using the on-device model (Apple's or a local LLM). If the answer confidence is low or the query is complex, it can be configured to automatically and seamlessly forward the same prompt to a cloud model (e.g., Google Gemini, Anthropic Claude) to generate a more comprehensive answer, all within the same interface.