Is my file sent to a server?

No. Only the model weights are downloaded. Your images and text are processed inside your browser.

Why is the first run slow?

The model has to download once. After that it loads from the browser cache in a few seconds.

My computer has no WebGPU. Does it still work?

Yes. The model runs on the CPU in a background worker. It is slower but the page stays responsive, and you can cancel at any time.

Are my files uploaded?

No. This tool runs entirely in your browser tab. Files are read from your disk into memory and the result is generated locally.

Is it free? Are there limits or watermarks?

Free and open source (MIT). There are no file-size caps, daily quotas, accounts or watermarks.

Does it work offline?

Yes, once the page (and any engine it needs) has loaded, it keeps working without a connection thanks to the service worker cache.

Home / AI / Semantic File Search

Semantic File Search

Search many PDFs, Word files, notes and code by meaning, not just exact words, all offline.

Free, no uploads, no sign-up. Runs 100% in your browser.

How Semantic File Search works

The model runs with Transformers.js and ONNX Runtime inside a background worker, so the page never freezes. It downloads once from the Hugging Face model hub and is then cached by your browser, so it works offline. With WebGPU it uses your graphics card, without it the CPU. Your files are never uploaded. Every passage is turned into a MiniLM sentence embedding, and results combine meaning with a small keyword boost. Keyword mode uses BM25 ranking with no download.

Input: PDF, DOCX, TXT, MD, CSV, EPUB, HTML. Output: CSV.

How to use Semantic File Search

Drop your documents.
Type a question or topic in plain words.
Press Search. The first search reads and indexes the files.
Open the matching passages or export them as CSV.

Tips

Ask the way you would ask a person, like "when is the rent due".

Semantic File Search FAQ

Is my file sent to a server?: No. Only the model weights are downloaded. Your images and text are processed inside your browser.
Why is the first run slow?: The model has to download once. After that it loads from the browser cache in a few seconds.
My computer has no WebGPU. Does it still work?: Yes. The model runs on the CPU in a background worker. It is slower but the page stays responsive, and you can cancel at any time.
Are my files uploaded?: No. This tool runs entirely in your browser tab. Files are read from your disk into memory and the result is generated locally.
Is it free? Are there limits or watermarks?: Free and open source (MIT). There are no file-size caps, daily quotas, accounts or watermarks.
Does it work offline?: Yes, once the page (and any engine it needs) has loaded, it keeps working without a connection thanks to the service worker cache.

Related tools

Search in PDFs: Full-text search across many PDFs at once, with context and the match highlighted on the page.
Chat with PDF: Ask questions about a PDF or document and get answers with page numbers, fully offline.
Name & Entity Extractor: Pull people, places, organisations, dates, money, emails and phone numbers out of any text or document.
AI Summarizer: Summarize PDFs, Word files and long text on your device, as paragraphs or key points.