Launching today
SynapseClean
Compact web copy into AI prompts — 100% on-device
12 followers
Compact web copy into AI prompts — 100% on-device
12 followers
SynapseClean intercepts copy and compacts messy webpage selections into clean Markdown outlines before they hit Claude, ChatGPT, or Gemini. Two-phase pipeline: rules strip boilerplate instantly with zero network calls; optional Gemini Nano refines in the background when Chrome's Prompt API status is available — often ~90% smaller on large pastes. Wait for the final toast before pasting when Gemini is refining. Triggers: auto-copy, Alt+Shift+C, context menu. MIT licensed. Free forever
Hi Product Hunt! 👋
If you research in the browser and paste into AI chats, you know the pain: you copy one article and your clipboard becomes a landfill — cookie banners, nav crumbs, newsletter CTAs, related-post spam, and footer legal text all hitch a ride into Claude or ChatGPT. One bloated paste can eat a whole premium message.
I got tired of manually cleaning every selection, so I built SynapseClean. It intercepts copy the moment you select text (or when you hit ⌥⇧C / Alt+Shift+C) and compacts it locally into a laser-focused prompt outline — often turning ~4,000 words into ~400.
What makes SynapseClean different:
Two-phase, 100% on-device — Rule-based compaction writes to your clipboard instantly with zero network calls. If Gemini Nano is enabled and Chrome's Prompt API status is available, a background pass refines that output for the tightest shrink (~90% on large pastes). Your text never leaves the browser. Wait for the final toast before pasting when Gemini is refining.
Zero friction — Auto-copy intercept, keyboard shortcut, or right-click context menu. Output lands on your clipboard ready to paste into Claude, ChatGPT, Gemini, Copilot, Perplexity, and more.
Real research workflows — Markdown, outline, or bullet output formats. Strips comment sections, ads, and invisible HTML artifacts.
Open source forever — MIT licensed at github.com/fujiDevv/synapseclean. No Pro tier, no license keys, no accounts.
I built this for researchers, writers, students, and anyone who lives in 30 tabs and 5 AI chats. Same privacy-first philosophy as my other project Arcrawls — local AI that actually respects your data.
Would love your feedback: what sites produce the worst copy-paste garbage for you? And what output format (outline vs bullets vs Markdown) do you reach for most?
Happy compacting!
How does the rules-based stripping handle sites with heavy dynamic content or paywalled articles where boilerplate is hard to define statically?
@melihcatallx79 SynapseClean bypasses paywalls and dynamic code by only processing the text you actively highlight on your screen, rather than scraping the website's underlying HTML. Once captured, the engine ignores HTML tags and evaluates the text's structure line-by-line using structural heuristics. It automatically drops blocks with a high density of short links or bullet points, assuming they are navigation menus or dynamic sidebars. Additionally, it deduplicates repeating lines to remove sticky headers and uses text-matching to strip out common dynamic injections like cookie banners, newsletter pop-ups, and comment threads.
Curious if the Gemini Nano refinement respects site-specific quirks like code blocks or tables, or does the rules phase occasionally strip out stuff it shouldn't on technical docs?