One search box for your library and the open web
Searching your own papers and discovering new ones are two different jobs. Here is how ScholarThread splits search into just two scopes — Local and Online — and why that beats a wall of filters.
Most reference tools make you decide, before you type a word, which kind of search you want. Titles or full text? Notes or papers? This tag or that folder? By the time you have set the filters, you have forgotten what you were looking for. Search should meet you where you are, not interrogate you first.
We recently collapsed search in ScholarThread down to two scopes, and only two: Local and Online. Everything you might want to find falls cleanly into one of them, and the split maps onto a real difference in what you are trying to do.
The two jobs of "search"
When a researcher types into a search box, they are doing one of two things:
- Finding something they already have. A paper they saved last month, a note they wrote during a reading session, a half-remembered passage in a PDF. The answer is already in their library; the problem is retrieval.
- Finding something they do not have yet. A new paper on a topic, work that cites a key result, the next thing to read. The answer is out in the literature; the problem is discovery.
These are different jobs with different sources, and conflating them is what produces the filter soup. So we stopped conflating them.
Local — everything you have, in one ranked list
The Local scope searches your whole library at once: your saved papers, the notes you have written, and the full text of the documents you have read. You do not choose "search notes" versus "search papers" — you search, and the most relevant things come back regardless of what type they are. A note you wrote, the paper it is attached to, and a passage from a third paper can all appear in the same ranked list, because all three might be the right answer.
Crucially, Local search is not just keyword matching. It combines two ways of finding things:
- Keywords, for when you remember the exact term — an author name, a method, a phrase you know is in there.
- Meaning, for when you do not. If you search "models that weight some inputs more than others" and a note talks about attention mechanisms without ever using your words, meaning-based search still surfaces it.
The two are fused into a single list, so you get the precision of keyword search and the recall of semantic search without having to pick. We wrote about the meaning half of this — and why, for a lot of people, it runs entirely on their own device — in a companion post on on-device semantic search.
Online — the open literature, when your library runs out
The Online scope is for discovery. When the answer is not in your library yet, ScholarThread searches the open scholarly record — hundreds of millions of works — so you can find new papers by title, author, or topic and pull the relevant ones straight into your library. It is the same box, one toggle away, so moving from "what do I already have on this?" to "what else is out there?" is a single click rather than a context switch to another tab.
Why two scopes and not ten
It is tempting to expose every axis a power user might want: separate tabs for papers, notes, highlights, collections, tags. We had a version that leaned that way, and the lesson was clear — every extra scope is a decision the user has to make before they know what they will find. Two scopes removes almost all of that up-front cost:
- You never have to guess which sub-search hides the thing you want. Local looks everywhere you have been.
- The distinction that remains — mine versus not-yet-mine — is one you already hold in your head when you start typing.
- Ranking does the sorting that filters used to do. Instead of narrowing to "notes only," you just search, and your notes rank where they deserve to.
Fewer choices, better defaults. The machine should absorb the complexity, not hand it to you as a settings panel.
A worked example
Say you are writing a discussion section and you half-remember reading something that pushed back on a result you are citing. You do not remember the author, and you certainly did not tag it "counterargument."
- In Local, you type the idea in your own words — "evidence against the recovery effect". Meaning-based ranking surfaces the paper whose abstract frames it differently, and the margin note where you flagged the tension weeks ago. One search, both halves of the memory.
- The note reminds you the critique was partial, so you want the follow-up work. You flip to Online, search the topic, and the newer papers that extend the debate come back — ready to add to your library and read next.
Two scopes carried you from a vague memory to a cited paragraph and a reading list, without a single filter menu.
Doing this in ScholarThread
Open search and you will see one field and two scopes. Start in Local — it searches your papers, notes, and document text together, by keyword and by meaning, and ranks the results as one list. When you have exhausted what you own, switch to Online to search the open literature and bring new work in.
The goal was never a more powerful search UI. It was less of one: a box that understands what you mean, knows the difference between your shelf and the wider world, and otherwise gets out of the way.