Skim Recap is a Chrome extension that detects fast scrolling and summarizes the passage you skipped, helping readers recover context without leaving the page. Its summaries run locally in the browser with Gemma 4 E4B through LiteRT-LM.
Skim Recap is a Chrome extension that helps readers recover context after moving quickly through a web page. It detects a fast scroll, estimates the passage between the previous and current reading positions, and displays a short recap beside the cursor instead of summarizing the whole page.
The recap is generated locally in Chrome with Gemma 4 E4B Instruct through LiteRT-LM and WebGPU. The model is downloaded once and cached by the browser for reuse during the session, while the skipped article text is processed on-device rather than sent to a hosted LLM API. The workflow is intended for long articles, documentation, essays, and research posts where readers want to skim without losing the thread.
The extension monitors scroll velocity and treats scrolling at approximately 1,800 pixels per second as a potential skipped-reading event. After a 180-millisecond settling period, it calculates the distance moved and identifies the relevant passage.
Skim Recap extracts text from the range the reader moved past rather than summarizing the entire page. This keeps the recap focused on the context that was skipped.
Gemma 4 E4B Instruct runs locally through LiteRT-LM and WebGPU in a Chrome offscreen document. The model is downloaded once, stored with the browser Cache API, and reused across webpage tabs during the browser session.
The recap appears near the cursor and streams into the same reading surface. The panel includes local model status, a copy action, a layout control, and a translation action.
Focus mode renders the generated recap as a paragraph, while Smart mode presents the same generated text as numbered points. Switching layouts reshapes the existing result and does not run the model again.
Readers can select a phrase on the page or inside the recap card to reveal a Feynman action and request an explanation. This explanation flow is separate from the normal recap and does not run unless the reader asks for it.
Move quickly through a lengthy article and use the recap to restore the context between two reading positions without returning to the earlier paragraphs manually.
Scan documentation for the relevant section while retaining a compact reminder of a skipped explanation, prerequisite, or transition in the material.
Use passage-specific recaps to follow an argument when reading speed changes, without replacing the source with a summary of the entire page.
When a skipped passage or recap refers to a term the page does not define, select the phrase and request a Feynman explanation rather than expecting the bounded recap to add outside context.
It summarizes the text positioned within the passage the reader skipped during a qualifying fast scroll. It does not summarize the entire page.
Gemma 4 E4B Instruct runs locally in Chrome through LiteRT-LM and WebGPU, inside a Chrome offscreen document. The site states that skipped article text is not sent to a hosted LLM API.
The documented trigger uses a scroll velocity of approximately 1,800 pixels per second, followed by a 180-millisecond settling window. The default skipped distance is 1,200 pixels, with an adjustable range of 400 to 3,000 pixels.
Not through the normal recap, which is intentionally limited to the skipped passage. A reader can select a phrase and use the Feynman action to request a separate explanation.
No. They use the same generated recap. Focus displays it as a paragraph, while Smart reshapes it into numbered points without regenerating the model output.
流量数据仅供参考。
noiz.io
AI YouTube 视频摘要工具，提取要点、问答和时间戳摘要
glarity.app
Glarity 使用 ChatGPT 式 AI 总结和翻译在线内容。
chatgptoutlookaddin.com
使用自定义提示词在 Outlook 中总结和撰写邮件的 AI 插件
usegenie.ai
基于网页的聊天机器人，支持多模型提示、PDF、数学和摘要工具
scisummary.com
为研究人员和学生总结、整理并比较科研论文的 AI 工具
pdfsummarizer.net
免费的 AI PDF 摘要工具，支持聊天、翻译和要点提取
