What Hushmark does
Hushmark helps you prepare documents before you share them with an LLM or another external service. It reads supported files in your browser, identifies possible personal or sensitive information, and lets you review those findings before you export a redacted copy.
Files stay on your device
In the browser version, file contents are read with browser APIs and are not uploaded to a Hushmark server. Document text is passed only between the page and the local worker used for detection. Remote model loading is disabled.
The macOS app packages the web assets, model, and WebAssembly runtime inside the app bundle. It does not require a Hushmark server for its local redaction workflow.
Local detection and exported files
Detection combines local patterns, a lightweight local NLP engine, and a quantized GLiNER model that runs through a local WebAssembly runtime. Detection is review-first, so you remain responsible for checking the document and deciding which findings to redact.
When you export a PDF, Hushmark rasterizes the redacted pages and checks for a surviving selectable text layer. DOCX, TXT, and HTML exports use replacement tokens. The resulting file is written through your browser download flow or, in the macOS app, to the location you choose in the native save panel.
Information this site may receive
This portfolio page does not provide a Hushmark upload or analysis service. A visit to shrikar.dev may produce ordinary technical data such as a request timestamp, IP address, browser details, and requested URL in infrastructure logs. No document content is collected by this page.
Your choices
You can use Hushmark without creating an account. You can review, deselect, or add detections before export. Do not treat automated detection as a guarantee. Manually review every file before sharing it.
Questions
For questions about these notes or the Hushmark MVP, email shrikar84@gmail.com. These notes will be revised if the product gains accounts, payments, or server-side document processing.