Inshō
印象 · impression — of everything you read
"I read something about this last week..."
Was it that blog post? That thread? That doc someone shared in Slack?
You open browser history. Scroll. Nothing.
Bookmarks? You never saved it.
You read 50+ pages a day.
You remember almost none of them.
What if you could just ask?
You ask:
What was that article about attention mechanisms I read last Tuesday?
You read "Attention is All You Need — 7 Years Later" on arxiv-digest.com last Tuesday around 2pm. You spent about 12 minutes on it.
The article argued that the original transformer architecture remains surprisingly unbeaten at scale, despite hundreds of proposed alternatives.
arxiv-digest.com · April 1, 2026 · 12 min read
Summarize the key points
From the article:
· Multi-head attention scales better than any proposed replacement
· Sparse attention variants sacrifice quality for marginal speed gains
· The real bottleneck moved from attention to memory bandwidth
· Next breakthrough likely in how context is stored, not computed
Compare this with the Mamba paper I read last month
You read "Mamba: Linear-Time Sequence Modeling" on March 5. Interestingly, that paper is exactly the kind of alternative the Tuesday article argues against. The attention paper claims Mamba's linear scaling comes at a quality cost that only shows up at 70B+ parameters...
Remember everything. Retrieve in any form.
Inshō quietly captures what you read and lets you retrieve it however you need — summaries, comparisons, exact quotes, or connections you didn't know existed.
Everything above happened on your machine. Here's what left your device:
Nothing.
Inshō runs locally. Your browsing history, your memory, your AI — all on your device. No cloud. No account required.
Inshō is a Chrome extension that captures and a desktop app that answers.
In beta · Launching April 2026 · More browsers and platforms coming soon.