Save the sentence, not the page. Highlight anything you read on the web, keep it in a searchable library, and ask questions about everything you've saved — with answers that cite the passages they came from.
Try it: askmarian.com · install from the Chrome Web Store
Marian is a Chrome extension for highlighting, noting and finding things again. You highlight any passage on any page, add notes, organise into collections, and search everything you've ever marked. Highlights are persistent — they survive page redesigns and edits rather than breaking the moment a site ships a new layout.
A passage is selected on an arbitrary page — including single-page apps whose DOM moves after load. The selection is anchored so it survives the page changing underneath it.
Bookmarks save the wrong thing. A URL tells you nothing about why you saved it — the sentence that made you stop and think is the part worth keeping, and that's exactly the part a bookmark throws away. Marian saves the passage, not the address.
The interesting part is what happens once you've saved enough. You can ask about a single passage and get an answer grounded in the page it came from — or ask your whole library "what do I know about this?" and get an answer that cites the specific passages it drew from, and says so when it doesn't know.
That last part is the one that matters. A retrieval system that will admit the answer isn't in your library is worth more than one that's right slightly more often but never tells you which time it isn't. The AI features are opt-in; everything else works without them.
No analytics, no tracking, no ads. It deliberately doesn't run on banking or healthcare pages, and works offline by default — an account is only needed for sync.
AI stack
Client
Deployed on
Avahana — our own deployment platform
Keeping retrieval and generation server-side is what makes library-wide questions possible at all — the extension alone can only see the page you're on. Qdrant holds the passage vectors, MongoDB the library and its metadata, and the agent layer is built with Strands so the model, tools and retrieval stay swappable rather than welded to one provider.
Live on the Chrome Web Store at v0.1.0.