Given, family, and full names across Latin and CJK scripts.
Privacy Filter Online
Detect names, emails, phone numbers, addresses, account numbers, dates, URLs, and secrets locally in your browser, then redact them in one click.
Model not loaded
The first run downloads model files from Hugging Face. Chrome or Edge with WebGPU is recommended.Input text
Results
0Labeled source
Waiting for detection.
Redacted text
Waiting for detection.
Image input
Run local OCR, then map detected privacy labels back onto the image.
Image OCR result
0Choose a demo image or upload your own image.
OCR debug
100% local Inference runs in your browser. No text is uploaded.
Open model Powered by the public openai/privacy-filter weights.
WebGPU + WASM Hardware-accelerated when available, with a WASM fallback.
Free forever No sign-up, no quota, no tracking — just paste and go.
How Privacy Filter works
- Load the model in your browser
The openai/privacy-filter weights are downloaded once from Hugging Face and cached for future visits.
- Run token classification locally
Transformers.js executes the model on WebGPU when available, with WebAssembly as a fallback.
- Review and copy
Inspect the highlighted entities, copy the redacted text, or paste new content — nothing ever leaves your device.
Supported entity types
Eight categories of personal data the model can label in any input text.
Personal and corporate email addresses, including country TLDs.
Mobile, landline, and international formats with separators.
Postal and street addresses, including ZIP codes and prefectures.
Customer IDs, bank accounts, vendor references, and order numbers.
Specific calendar dates that could identify a person or event.
Sensitive links such as private invoices or one-time session URLs.
API keys, access tokens, and other credential-like strings.
Use cases
Strip personal information before pasting into ChatGPT, Claude, Gemini, or any chatbot.
Redact customer data from support tickets, error logs, and bug reports before sharing.
Mask names, addresses, and account numbers before sending screenshots or text snippets.
Catch API keys and tokens accidentally embedded in documentation, README files, or chat threads.
Frequently asked questions
Is my text uploaded to a server? +
No. Inference runs entirely in your browser through Transformers.js. After the initial model download from Hugging Face, your text never leaves the device.
What kinds of personal information can it detect? +
Names, emails, phone numbers, addresses, account numbers, dates, URLs, and secrets such as API keys or tokens.
Which browsers are supported? +
Chrome and Edge with WebGPU offer the fastest performance. Other modern browsers fall back to WebAssembly automatically.
Why is the first detection slow? +
The first run downloads the openai/privacy-filter weights. Subsequent runs use the browser cache and are noticeably faster.
Is Privacy Filter free? +
Yes. The page is free to use, and the underlying openai/privacy-filter model is published with open weights on Hugging Face.
Can it replace a regex-based redactor? +
For most natural-language text, a contextual model catches entities that regex misses (multilingual names, free-form addresses) — though combining both gives the best coverage.