Chat With Your PDF
Drop in a PDF and ask it anything. A real language model and document search run entirely inside this browser tab on your GPU. Your file is never uploaded, there is no API key and no sign-up, and it keeps working offline once the models download.
🔒 100% private — your document never leaves your deviceDrop a PDF here
or click to choose a file · it is read on your device, never uploaded
How it works
Read & split
Your PDF is parsed in the browser and split into overlapping passages, each tagged with its page number. The text never leaves the tab.
Embed
A small embedding model turns every passage into a vector on your device, so the tool understands meaning, not just keywords.
Retrieve
When you ask something, your question is embedded and matched against the passages with a cosine vector search to find the most relevant ones.
Answer
A local language model reads only those passages and writes an answer that cites the pages it used. Click a page to read the source.
How does a chatbot read my PDF with no server?
This is retrieval-augmented generation, the same pattern used by hosted "chat with your documents" products, except it runs end to end on your own machine. Two open-source models are loaded straight into the page: a compact embedding model for semantic search and a quantized language model that runs on your graphics card through WebGPU. The document is chunked and embedded locally, the question is matched to the closest chunks, and the model is asked to answer using only that retrieved context and to cite page numbers. There is no backend, no queue, and no per-question cost, because all of the work happens in your browser.
Why run it locally?
Privacy is the point. Contracts, medical records, financial statements, legal filings, and unreleased drafts are exactly the kind of documents people are nervous about pasting into a cloud service. Here nothing is transmitted, so it is safe for confidential material. It also costs nothing per question, has no rate limits beyond your own hardware, and keeps working on a plane or anywhere offline once the models are cached. Document search works in any modern browser; written answers use a local model that needs WebGPU, and if that is unavailable the tool still surfaces the most relevant passages from your file.
Good things to try
Summarise a long report, pull a specific number or clause out of a contract, ask what a paper concluded, find every mention of a term, turn a policy document into plain-English answers, or quiz yourself on study notes. Because answers are grounded in retrieved passages and cite their pages, you can always click through and verify the source rather than trusting the model blindly.
Is my PDF uploaded anywhere?
No. The file is read, split, embedded and searched entirely in your browser. The text never touches a server. The only network requests are the one-time downloads of the open-source models from a public model host.
How accurate are the answers?
Answers are grounded in passages retrieved from your document and cite the pages they came from, which keeps them honest and checkable. The local models are small enough to run on consumer hardware, so for very long or subtle documents you should still click the cited pages to verify. If something is not in the document, the tool is told to say so rather than guess.
What do I need to run it?
Any modern browser can read a PDF and search it. Generating written answers uses a language model that needs WebGPU, available in recent desktop Chrome or Edge. On a browser without WebGPU the tool falls back to showing you the most relevant passages with their page numbers.
Does it really work offline?
Yes. Once the embedding and language models have downloaded, the browser caches them. You can reload, go offline, and keep asking questions about your documents.
How big are the downloads?
The embedding model is small, around 30 MB. The answer model is a few hundred megabytes to roughly a gigabyte depending on which one you pick. Each downloads once and is then reused from cache.
Is it free?
Completely free, with no account and no API key. Because it runs on your own machine there is nothing to bill.
I build tools like this every day.
Senior full-stack engineer, available for senior or contract work, fully remote. See the rest of the lab or get in touch.