Starting…
Search your documents without them leaving this computer.
Reading, embedding, indexing and searching all happen inside this browser tab. Nothing is uploaded, and the counter at the top proves it.
Index a folder
Pick a folder of text and Markdown files. The page builds a searchable index in this browser's own storage.
Try the sample
Twenty arXiv abstracts, indexed in seconds. A quick way to watch the whole pipeline run.
How it works
01ReadFiles are read from your disk and split into passages.
02EmbedMiniLM turns each passage into 384 numbers, on your GPU.
03IndexPassages are grouped into shards and indexed with LM-DiskANN.
04SearchOnly the shards closest to your question are searched.
Working
Starting…
waiting
Progress is saved as it goes. If this tab closes, choose the folder again: finished work comes back from the cache instead of being redone.
Your current index is kept until the new one is ready. Cancelling leaves it exactly as it was.
Show details
| Phase | Progress | Elapsed | Left | State |
|---|
Striped bars are estimated from a measured rate: building an index reports no progress of its own. Solid bars count real work done.
This index lives in the browser's own storage. Quit the browser completely and reopen this page: it opens the index straight away, with no re-indexing.