Find the essence
of any text.
Paste an article, a paper, a transcript. Saramsha scores every sentence by how much of the document's meaning flows through it, then keeps only the ones carrying the weight.
No account, no upload, no API. Everything runs in this browser tab.
Source
सारांश
Recurring terms
How it works
Importance isn't a property of a sentence.
It's a property of its neighbours.
A line that would be trivial in one document can be the hinge of another. So Saramsha never reads a sentence alone. It treats the whole text as a network: every sentence becomes a node, and every pair of sentences is joined by an edge weighted by how much meaningful vocabulary they share.
Then it asks a recursive question — which sentences resemble the sentences that themselves resemble many others? Weight flows around the graph and settles, after enough iterations, into a stable ranking. This is TextRank, built on the same insight that once ranked the early web.
The approach has an appealing honesty. It invents nothing, cannot paraphrase, and will never produce a sentence you didn't write. What it gives you instead is a defensible claim about emphasis: not what the document should have said, but what it actually kept saying.