RedactPDF

Redact PDFs on your own machine. Mark what has to go (a passage of text, an image, a whole page, the phone numbers a detector picks out) and get back a copy with that content removed from the document, not hidden under a black rectangle.

Nothing is uploaded. No network calls, no account, no telemetry. Download a binary for Windows or Linux, or pipx install blankout on any machine that has Python. Nothing else to install, text recognition for images included.

demo.mp4

_{Content is deleted from the file, not covered with a black box. Check it yourself with a different PDF library: Check the output yourself.}

---

What it does

You open a PDF in your browser, mark content in one of five ways, and export:

Way in	What it covers
Select text	highlight it in the viewer, redact the selection
Draw a rectangle	whatever the text layer cannot reach: a scan, a signature, a logo
Censor a page	wipes the page whole, images and vector graphics included
Type a pattern	an exact string or a regular expression, with options for word boundaries, accents and line breaks
Pick a detector	e-mail addresses, phone numbers validated with libphonenumber, card numbers filtered by Luhn

Whatever you mark is deleted from the file. Images are handled as well: by default only the pixels you targeted are blackened inside the bitmap itself, which is what makes a flattened scan redactable without losing the rest of the page.

Every export is checked before you get it

After redacting, RedactPDF re-runs the rules you gave it against the file it just produced. If anything you targeted is still findable in there, you do not get the file. You get a report naming what survived, on which page, and whether the engine was able to locate it at all.

There is a third answer, and it matters more than it looks. An audit can only report that a target survived; it can never report that it was unable to look. A name printed inside a scan is invisible to the rules and equally invisible to the check, so the export used to succeed with the name plainly on screen. RedactPDF now says so instead: an area it could not read stops the export and is shown to you, one by one, to confirm. You can turn that off, but only by naming the choice.

That check has caught real failures, and it has blind spots of its own. Both are written down in docs/SECURITY.md rather than left implied; read it before trusting the tool with anything that matters. And if you would rather not take its word for it, three commands from a different PDF library will tell you independently: see Check the output yourself.

---

Download

Take the file for your system from the latest release and run it. No Python, no Node, nothing to install. The app opens in your browser and stops when you close the tab.

System	File
Windows	redactpdf-windows-x86_64.exe
Linux (x86-64)	redactpdf-linux-x86_64, chmod +x it first

Windows shows "Windows protected your PC" on first run, because the executable is not signed with a paid code-signing certificate. Click More info, then Run anyway. docs/WINDOWS_BUILD.md explains why. On Linux the binary needs a glibc at least as recent as Ubuntu 22.04's.

Since the binary is unsigned, the release page gives you two ways to check it yourself instead. SHA256SUMS.txt says the bytes are the ones that were built:

sha256sum --ignore-missing -c SHA256SUMS.txt

And the build attestation says this repository built them, at a known commit, on GitHub's runners, which a hash alone cannot tell you:

gh attestation verify redactpdf-linux-x86_64 --repo theodubus/RedactPDF

With Python: any OS, and the recommended path on macOS

pipx install blankout
blankout

The package is called blankout because the name that matches this repository was already too close to an existing one on PyPI. Installing it also puts redactpdf on your PATH, which is the same command under the name the downloadable binary carries; use whichever you remember.

Same app, same single command to run it. There is no macOS binary: a file downloaded through a browser carries a quarantine attribute, and since macOS Sequoia clearing it takes a trip through System Settings and an admin password. pipx sidesteps that entirely, because nothing is downloaded by the browser and Gatekeeper never applies. It also makes the install independent of the build machine's glibc, which is what constrains the Linux binary above.

uv tool install blankout does the same job. Plain pip install works inside a virtual environment you manage yourself, but into a system Python most current Linux distributions refuse it outright (PEP 668).

From source

git clone https://github.com/theodubus/RedactPDF.git
cd RedactPDF/frontend && npm ci && npm run build && cd ..
cd backend && python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]" && cd ..
python launch.py

---

Documentation

Document	What is in it
Usage	The five kinds of rule, the three image modes, what the matching does that the UI does not show, and how to verify an export with software that is not this project.
Security model	What is guaranteed, what is not, and the known limits, including the ones that are structural rather than bugs. Also covers exposing the API beyond a single local user.
Development	Running the two halves with hot reload, the test suite and its PDF fixtures, building the binary and the wheel, and what CI checks.
Windows build	The failure modes of the frozen Windows build, most of which produce no output at all.

---

License

RedactPDF is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). See LICENSE.md for the full text.

In short: you may use, study, modify, and redistribute it freely, but any redistributed or network-hosted (SaaS) version, including modified ones, must make its complete corresponding source available under the same AGPL-3.0 terms. This copyleft is also required by the core dependency PyMuPDF, which is itself AGPL-3.0 (or commercial).