Name: Skim Recap
Rating: 4.0 (1 reviews)

Hi Product Hunt 👋 I built Skim Recap because I kept flicking past several screens of long articles, then wondering whether I had skipped the part that mattered. It detects a fast scroll and recaps exactly that skipped stretch beside your cursor—not the whole article and not everything above it. The feature I’m happiest with in 0.4 is Feynman. A recap is deliberately limited to what the passage said. That keeps it faithful, but it also means a recap cannot explain a term the page never defined. My first attempt simply re-explained the whole passage. It produced a longer recap, because the extension still did not know which word had stopped me. Now you select a word or phrase—on the page or inside the recap card—and press Feynman. The selection becomes the subject. Gemma uses its paragraph, nearby context, the closest heading, and the page title to decide which meaning fits. “Argument” in a logic article and in a function signature should not receive the same explanation. Also new in 0.4: recaps stop at a sensible number of points, Retry asks for a genuinely different expression, and the card runs in a shadow root so host-page CSS cannot restyle it. Everything runs locally with Gemma 4 E4B, LiteRT-LM, and WebGPU. No account or hosted LLM API. The honest trade-off is a one-time 2.97 GB model download, plus WebGPU and substantial local disk/memory requirements. I’d especially love feedback on two things: does the fast-scroll threshold feel natural, and does the Feynman explanation use enough context without becoming too long?

Skim Recap

Recaps what you skipped and explains where you get stuck

Recaps what you skipped and explains where you get stuck