redact-paperasse-demo

A small public demo of redact-paperasse — upload plain text, a PNG/JPEG image, or a PDF, pick which kinds of PII to redact, and get a redacted copy back.

Live at redact.paperasse.ai.

How it works

• Backend: Hono on Node 22, one flat router (src/app.js). POST /redact takes a multipart upload + a JSON entities array, dispatches to redactText/redactImage/redactPdf from the redact-paperasse npm package by MIME type, and streams the redacted bytes straight back as the response body.
• No files ever touch disk. The upload is read into memory, redacted in memory, and served back in the same request — "delete after download" is true by construction, not a cleanup step working against a file that was ever written.
• Frontend: a single static page (public/index.html), vanilla JS, no build step.
• Counter: data/stats.json, incremented per successful redaction, written via an atomic temp-file-then-rename swap so a crash mid-write can't corrupt it. The UI only shows the running count once it passes 1000.
• Abuse protection: a 10MB upload cap, a MIME-type allowlist, a simple per-IP rate limit, and a concurrency cap on simultaneous redactions (OCR/PDF rendering is the expensive part, not the HTTP handling).

Supported entity types

Whatever redact-paperasse currently registers as Tier A recognizers: email addresses, phone numbers, credit card numbers, IBANs, US Social Security Numbers, and French social security numbers (NIR). See src/redact.js for the exact list — the library doesn't expose a "list supported entities" call today, so this is hand-kept in sync with crates/recognizers/src/lib.rs.

Local development

npm install
cp .env.example .env
npm run dev

Then open http://localhost:8790.

Deployment

Deployment configuration (server details, hostnames) is intentionally not included in this repo — it's environment-specific and kept out of version control.