SanitizeGPTGet the free beta ↗

WINDOWS BETA · FREE EARLY ACCESS

Your documents.
Your computer.
Your call.

Review sensitive information locally before you bring AI into the conversation.

SanitizeGPT finds sensitive details in text-based PDFs, lets you decide what to share, and shows you the sanitized context before optional cloud chat.

No card. Free during beta. Your feedback shapes what comes next.

LOCAL REVIEW On your computer

Client brief Fictional example

Please contact Alex Morgan at alex@example.test about the project timeline.

2 details to reviewHuman review required

Outbound preview Sanitized

Please contact [PERSON_1] at [EMAIL_1] about the project timeline.

↳ Original values stay in the local session.
Local PDF processingQwen detection via OllamaReview before sharingOptional cloud chat

A MOMENT OF REVIEW. A LOT MORE CONTROL.

From document to conversation,
with you in the middle.

Automated detection starts the work. You make the final decision.

01 / OPEN

Keep the source local.

Open a text-based PDF. Extraction and sensitive-information detection run on your computer using the local Qwen model.

02 / REVIEW

Decide what is shared.

Redact a detail, allow it, or block the conversation. Consistent placeholders keep repeated entities connected within your chat.

03 / ASK

Check the actual preview.

Review the sanitized documents and prompt before sending to your chosen cloud AI. Original values are restored locally in the answer.

TRY A FICTIONAL EXAMPLE

One detail. Your decision.

“Please ask Alex Morgan to confirm the timeline.”

OUTBOUND PREVIEW

Please ask [PERSON_1] to confirm the timeline.

Redacted: a placeholder would be sent. This demo sends nothing.

A LOCAL APP. A FAMILIAR BROWSER.

A workspace for
careful questions.

The installer starts SanitizeGPT on your computer and opens its interface in your browser. Your local service listens on loopback.

Review documents, manage sensitive details, and inspect what will reach the AI provider in one place.

Read the illustrated installation guide ↗
SanitizeGPT local browser interface showing a fictional document and chat example
Actual Windows beta interface · fictional data and a local test reply

HELP SHAPE THE NEXT VERSION

Early days.
Useful feedback.
Free beta.

Try the workflow and tell us where it helps, where it misses, and what gets in your way.

  • Windows x64 · text-layer PDFs
  • Free during beta; no announced end date
  • No payment card or automatic billing

Early beta: unsigned installer. Detection can miss details.
Clean-machine verification is still pending.

YOUR INVITATION TO TRY IT

Get the free Windows beta.

Verify your email to unlock the installer. No password to remember.

Checking beta availability…

By requesting access, you acknowledge the beta notes and website privacy notice. Updates are optional; you can ask us to delete your account.

BEFORE YOUR FIRST DOCUMENT

A short setup.
A local model.

What you need

  • Windows 11 or Windows 10 22H2+, x64
  • 16 GB RAM recommended
  • About 12 GB free disk space for setup and model
  • Internet for verification and initial downloads
  • Your own OpenAI API key if you enable cloud chat

Python and Node.js are included or unnecessary. Ollama and model weights are separate from the installer.

1

Verify your email and install.

Download the Windows installer and follow the illustrated guide. Installation is per user by default.

2

Set up Ollama and the Qwen model.

Install Ollama if needed. SanitizeGPT reuses a compatible local model or asks before downloading the pinned qwen3.5:4b model (about 3.32 GB).

3

Start with a fictional PDF.

Review every finding and the outbound preview. Scanned PDFs and OCR are outside this beta.

Installation guide with screenshots ↗Installer checksum ↗

GOOD QUESTIONS

A little clarity
before you start.

Still have a question?
hello@sanitizegpt.com

Is the beta free?

Yes. SanitizeGPT is free during beta, with no announced end date. No payment card or automatic billing. Future pricing will be announced separately.

Do my original documents leave my computer?

PDF extraction, Qwen detection, review, and placeholder mappings run locally. Optional cloud chat sends the sanitized context and prompt that you review. Detection can miss information: always inspect the outbound preview.

What can I open?

The Windows beta supports PDFs with an existing text layer. Scanned PDFs and OCR are not supported. Review up to five documents in a chat session.

What is included in the installer?

The installer includes the browser interface, Python processing runtime, and self-contained launcher. You do not need Python or Node.js. Ollama is installed separately; the pinned qwen3.5:4b model is downloaded only with your permission.

Can I use it offline?

After Ollama and the model are ready, local PDF review and detection can work offline. Cloud chat requires internet access and your own supported OpenAI API key. Initial downloads and email verification also require internet access.

What should I know before trying it?

This is an early Windows x64 beta. The installer is unsigned, detection needs human review, and clean-machine verification is still pending. Use synthetic or de-identified documents for your first test.