Step 2
Ask
Only your question and the few most relevant passages (shown on the right) are sent to the model. The document stays in your browser.
Free AI tool
Ask your document
A contract, a tender, a 60-page report: drop it in and ask. The file is read and indexed in your browser; for each question only the handful of relevant passages is sent to the model, and the answer cites them so you can check. This is a one-document version of how our KnowledgeRAG platform works.
Step 1
Your document
Drop a PDF, Word or text file here — or paste text below. It is parsed and indexed in your browser; the file is never uploaded.
checking…
Answers
Conversation
Load a document and ask a question. Answers cite passages as [1], [2]…
How it works
Retrieval-augmented answers, kept small on purpose
1 · Parse locally
PDF text is extracted with pdf.js, Word with mammoth, plain text as-is — all inside the page. Scanned PDFs without a text layer are not supported (no OCR in the browser version).
2 · Chunk & index locally
The text is split into ~200-word passages with page numbers and indexed with BM25, the classic search ranking function. Nothing has left your machine yet.
3 · Retrieve, then ask
Your question ranks the passages; the top six are sent with the question to the model, which must answer only from them and cite [n]. Passages are shown next to the answer.
Why not send the whole file?
Privacy, cost and quality: models answer better from a few relevant passages than from 60 pages, and you keep control of what leaves the browser.
Languages
Ask in French, Dutch or English; the answer follows the language of the question. Documents can be in any of them.
Limits of the free version
One document at a time, keyword retrieval (no embeddings), a fair-use cap on questions per hour, no history kept between visits. The full platform lifts all of these.
Questions
What exactly is sent to the AI model?
Your question and up to six passages (~1,200 words) that the local search selected — you can see them in the “Passages used” panel. Requests go through OpenRouter to the model shown in the badge; we do not store or log content on our side.
Can it be wrong?
Yes. The model is instructed to answer only from the passages and to say when they do not contain the answer, and every claim is cited so you can verify in one click — but always check the source passage before relying on it.
Why did it miss something that is in the document?
Keyword retrieval only sees passages that share words with your question. Rephrase with the document’s own vocabulary (e.g. “termination” rather than “ending the deal”), or ask for a section explicitly.
Is there a size limit?
Files up to 25 MB; several hundred pages of text are fine. Very large files just take a few seconds to parse. Scanned image-only PDFs need OCR, which the browser version does not do.
Can I use it on confidential documents?
The document itself never leaves your browser. The retrieved passages do go to a third-party model provider for the answer — for genuinely sensitive material, redact first with our PII Redactor, or ask us about an on-premise deployment.
I want this over my whole knowledge base.
That is KnowledgeRAG: many documents, connectors, semantic search with embeddings, access control, on your infrastructure. Talk to us.
Same idea, all your documents
Semantic retrieval, connectors to SharePoint/Drive/Confluence, permissions, audit trail — deployed on your cloud or on-premise.