PROCESSED LOCALLY
⬡ zero uploads

Every PDF tool you need,
running inside your browser.

Compress, convert, merge, sign, redact and read scans. Choose a tool and open a file — it is processed on this device and never uploaded.

Why this is different

No document is uploaded. This can be independently verified.

Comparable services transmit your file to a server and undertake to delete it subsequently. PaperJET has no server to which a document could be sent: the tools are compiled to WebAssembly and execute within this browser tab, on your own device. Nothing is queued, nothing is capped, and the work continues if you go offline after the page has loaded.

To verify Load this page, disconnect from the network, and then compress a document. The operation completes as normal. No server-dependent service could do so.
No uploadThe file is read from local storage by the browser and does not leave it. No copy exists anywhere other than on your own device.
No account requiredNo registration, no email address and no consent banner. An account is optional and serves only to retain your preferences.
No limitsNo file size cap, no daily quota, no queue and no watermark applied to output. The only constraint is the memory of your device.
Established componentsBuilt on pdf-lib, pdf.js and Tesseract, the same libraries used across the industry, executed locally.

Questions

Frequently asked questions

Are documents uploaded to a server?

No. Every tool executes within your own browser using WebAssembly. The file is read from local storage by the browser and is not transmitted, so no server holds a copy of it. This may be verified by disconnecting from the network once the page has loaded; the tools continue to function normally.

Is PaperJET free to use?

Yes. All tools are provided free of charge, with no account required, no watermark applied and no file size limit.

Can a scanned PDF be processed?

Yes. The OCR tool reads the text from a scan and can return the same scan with an invisible text layer applied, so that the document appears unchanged while becoming searchable and available for copying.

Can a PDF be converted to Word or Excel?

Yes, where the PDF already contains text. Where the file is a scan, the OCR tool should be used instead; the interface will indicate which case applies.

Is there a file size limit?

No. As processing is performed on your own device rather than on a server, the constraint is the available memory of that device rather than an upload cap.

Compress
Drop a PDF here, or click to choose
It stays on this computer

Read the text from a scanned Marathi document

Marathi documents are scanned far more often than they are recognised, largely because most tools offer Hindi and leave it at that. The two share the Devanagari script but not the vocabulary, and a Hindi model reading Marathi makes predictable mistakes. This uses the Marathi model, and runs it on your own device rather than on a server.

How to read the text from a scanned Marathi document

  1. Open the scanned PDF. Marathi is already selected as the language.
  2. Run the recognition. The Marathi model downloads once on first use and is cached from then on.
  3. Save. The page keeps its original appearance, with the recognised text added as an invisible layer.
Why the language choice matters

Recognition is not only shape matching — the model weighs which words are plausible. A Hindi model asked to read Marathi will resolve ambiguous marks towards Hindi words that the document does not contain, which is why the same scan reads better under the right language.

Nothing leaves the device

Land records, school certificates and legal papers are the documents most often scanned in Marathi, and they are exactly the ones worth not uploading. The recognition happens in this tab.

Straight scans read best

Devanagari's headline and matras are thin, and they are the first thing lost to a skewed or poorly lit capture. A flatbed scan at 300 DPI gives markedly better results than a photograph.

Questions about Marathi OCR

Is my Marathi document uploaded?

No. Both the recognition engine and the Marathi model run locally in this browser tab. Nothing is transmitted.

Why not just use Hindi OCR for Marathi?

The script is the same but the vocabulary is not, and the model uses word plausibility to resolve unclear marks. A Hindi model reading Marathi will produce Hindi-shaped errors.

Does the original scan change?

No. The recognised text is applied underneath the existing page image, so the document looks identical and nothing is lost where recognition was imperfect.

Can it read handwritten Marathi?

Not reliably. The models are trained on printed text. Handwriting, particularly cursive Devanagari, is beyond what this can do.