Skim Recap

Product Introduction

1. Definition: Skim Recap is a privacy-first, on-device AI Chrome extension for reading comprehension and research acceleration. It is a technical tool that leverages local large language models (LLMs) to provide contextual summaries and explanations directly within the browser.
2. Core Value Proposition: It exists to solve information overload and comprehension gaps during fast-paced reading. Its primary value is enabling efficient skimming of long-form content—like articles, documentation, and research posts—without missing key points, while guaranteeing complete data privacy as all processing occurs locally on the user's GPU.

Main Features

1. Automatic Scroll-Based Summarization: The extension uses fast scroll detection algorithms to identify passages a user likely skipped. It then isolates that specific text segment and generates a concise, contextual summary using the Gemma 4 E4B model. This on-device summarization via the LiteRT-LM runtime ensures no reading data is transmitted externally.
2. Cursor-Native Recap Interface: Generated summaries (recaps) are displayed in a dynamic panel anchored near the user's cursor, minimizing context switching and eye movement. The interface includes a "Focus" (paragraph) and "Smart" (bulleted points) layout toggle, which reformats the same generated text without requiring a costly re-run of the local LLM.
3. Feynman Explanation Mode (Context-Aware Q&A): This feature allows users to select any term or phrase on the webpage or within a recap card. Activating the "Feynman" pill triggers the local Gemma model to generate an explanation. It synthesizes context from the surrounding paragraph, nearby content, heading, and page title to define the concept specifically as it's used on the page, going beyond a generic dictionary definition.

Problems Solved

1. Pain Point: The inefficiency and risk of missing critical information when speed-reading or skimming complex online content. Traditional methods force a choice between slow, thorough reading or fast, incomplete comprehension.
2. Target Audience: Technical professionals (developers, researchers, engineers), students, academics, journalists, and avid consumers of long-form digital content like documentation, essays, and investigative reports.
3. Use Cases: Quickly grasping the key arguments of a lengthy research paper, efficiently parsing API or software documentation, catching up on missed details when revisiting a partially-read article, and understanding domain-specific jargon without breaking reading flow to conduct external searches.

Unique Advantages

1. Differentiation: Unlike cloud-based summarization tools or browser assistants that rely on external APIs, Skim Recap operates entirely locally using WebGPU acceleration. This provides superior privacy (zero data leakage), eliminates latency and API costs, and allows functionality offline or on paywalled content.
2. Key Innovation: The integration of the lightweight, efficient LiteRT-LM runtime with the powerful Gemma 4 E4B model for local browser execution is a significant technical achievement. Furthermore, its "Feynman" feature represents an innovative approach to in-context Q&A, using the document's own structure and content to ground explanations, making it more accurate than a detached chatbot.

Frequently Asked Questions (FAQ)

1. How does Skim Recap protect my privacy while summarizing text? Skim Recap ensures maximum privacy by performing all AI processing locally on your device. The Gemma 4 LLM runs entirely within your browser via WebGPU and the LiteRT-LM runtime. No text from the pages you visit, your summaries, or your queries are ever sent to an external server or cloud API.
2. What are the system requirements to run Skim Recap's local AI model? Skim Recap requires a Chrome-based browser with WebGPU support enabled and a compatible GPU. The Gemma 4 E4B model is downloaded and cached locally once, requiring initial bandwidth and sufficient local storage. Performance scales with your device's graphical processing capabilities.
3. Can Skim Recap summarize any webpage or PDF? The extension is designed primarily for text-heavy articles, blogs, and documentation pages. Its effectiveness can vary on highly dynamic web apps, paywalled content (where it works locally on the rendered text), or complex PDFs viewed in-browser. It functions on any page where text selection is possible.
4. What is the difference between the "Focus" and "Smart" recap layouts? Both layouts present the exact same AI-generated summary text. The "Focus" layout displays it as a single, cohesive paragraph for fluent reading. The "Smart" layout intelligently reformats the same text into numbered bullet points for easier scanning of key ideas. Switching layouts is instant and does not re-trigger the AI model.
5. How does the Feynman explanation feature work without searching the web? When you select a term, the Feynman feature does not perform a web search. Instead, it instructs the local Gemma model to analyze the context you provided—including the selected text, the surrounding sentences, the section heading, and the page title—to infer and explain the term's meaning specific to that document.