Privacy & Data Use

Effective September 12, 2026

Browser-local document processing

PDF and image decoding, text detection, recognition, review, and PDF reconstruction run on your device. Typsastra does not upload source documents, full pages, filenames, paths, document hashes, OCR results, review decisions, or exported files as part of normal processing.

Browser storage and model caching

The site stores your light or dark theme choice and INT8 OCR thread count in browser local storage. OCR models, vocabulary, fonts, runtime files, and other application assets may be retained in the browser's HTTP cache. Content hashes allow unchanged models to be reused and cause a new download when a model version changes. This local data is not used to identify you and can be removed through your browser's site-data controls.

General feedback

The Feedback form sends only a schema version and the message you type. It does not attach document content, OCR text, filenames, page state, theme choice, or a user identifier. The application database stores a random record ID, the trimmed message, and its creation time.

A connecting network address is used transiently by Cloudflare's rate limiter to reduce abuse, but Typsastra does not write it to the feedback table. Feedback is anonymous and retained until reviewed or manually deleted. Because no account or contact information is attached, a particular submission generally cannot be linked back to its author later. Do not include private, confidential, or identifying information.

Optional reviewed-sample donation

A donation occurs only after you review an eligible sample, inspect the exact crop and text that will be sent, and select an unchecked consent box. The donation contains that displayed grayscale crop, raw and confirmed text, diagnostics, and relevant model and crop-setting versions. It does not contain the source document, full page, filename, source path, document hash, page or line identifier, account, cookie, or browser fingerprint.

Donated crops are stored privately in Cloudflare R2 and validated metadata is stored in D1 for OCR training, evaluation, and improvement under the permission shown in the donation dialog.

Aggregate analytics

The site records UTC daily totals for a fixed set of events such as page views, OCR starts, exports, donation prompts, rejected detections, and clicks on the five named model, segmenter, dictionary, and developer resource links. Link events contain only the fixed link category, not an arbitrary destination URL or referrer. These requests do not include document content, OCR text, filenames, geometry, confidence values, theme choice, cookies, or persistent visitor identifiers. The database stores only the day, event name, and aggregate count; it does not attempt to count unique visitors.

Infrastructure and external resources

The site and its APIs use Cloudflare Workers, Static Assets, D1, R2, rate limiting, and aggregate operational metrics. Browsers also request pinned PDF and ONNX runtime files from jsDelivr. Like other web infrastructure providers, Cloudflare and jsDelivr may process ordinary connection information under their own policies. Typsastra does not copy that operational information into its document, feedback, donation, or analytics records.

Third-party links and changes

Links to model, dictionary, source-code, or developer sites are governed by those sites' policies. This notice may be updated as the service changes; the effective date above identifies the current version.

Questions

Use the site's Feedback form for privacy or data-use questions. Do not include confidential information in the message.