Semantic Search
Paste a document and ask a question in plain language — the tool finds the paragraphs that match the meaning, not just the keywords. Runs a small AI embeddings model entirely in your browser.
Your data never leaves your deviceFrequently Asked Questions
-
How is this different from Ctrl+F?
Ctrl+F matches exact words or substrings. Semantic search matches meaning. If you ask "What is the deadline?" it can surface a paragraph that says "We must deliver by Friday," even though the word "deadline" never appears.
-
Which model is used?
MiniLM (Xenova/all-MiniLM-L6-v2), a 22 MB sentence-embedding model. Each paragraph and your query are converted to a 384-dimension vector; the tool ranks paragraphs by cosine similarity. Everything runs in your browser via @huggingface/transformers.
-
Is my document uploaded anywhere?
No. The model and all computations stay in your browser. The text never leaves your device.
-
What kinds of documents work best?
Anything you can copy-paste — meeting transcripts, policy documents, articles, chat logs, manual excerpts. Split long content into short paragraphs (a blank line between them works great) to get more precise matches.