parse▰
Read PDFs, Markdown, and plain text in the browser.
private document search · browser-first
Searchthe filesyou meant.
Drop in PDFs, Markdown, and plain text. Search with the words you know—or the meaning you remember. Nothing uploads. Nothing leaves the browser.
- no upload step
- no account
- local index
The useful difference
Search does not have to mean upload.
SearchThing treats your browser as the workspace: parse, chunk, embed, store, and rank without sending a copy of your document set to a hosted search service.
chunk▰
Keep the index useful instead of making one huge block.
store▰
Persist documents in IndexedDB and vectors in OPFS-backed storage.
rank▰
Fuse lexical and semantic matches into one readable result list.
Three ways to ask
Use the kind of search the moment calls for.
hybrid
Words + meaning
BM25 catches the exact phrase. Local vectors catch the idea you remembered differently. Reciprocal rank fusion brings them together.
vector
Search by the shape of a thought
Local embeddings help a query like “doctor visit” find a note that says “medical appointment.”
text
Instant and model-free
Text search works from the first keystroke, before the embedding model has finished warming up.
A smaller trust boundary
Your documents do not need a commute.
Files are read into browser storage and embeddings are generated by a web worker. SearchThing has no upload step in its working flow, so the private material stays in the place you opened it.
01 / StorageIndexedDB + OPFS-backed vectors
02 / FeedbackSee ranking signals instead of trusting a mystery score
Try the local index