100% Private
Runs fully in your browser. Nothing sent to any server. Ever.
No Signup
No account, no API keys. Open and use it instantly.
Free Forever
Permanently free. No trials, no limits, no credit card.
One of 64 free AI tools built by Zalt, an AI architect and software engineer.
Free PII Redactor (Safe-Paste for ChatGPT)
A privacy-first redaction tool that scrubs personally identifiable information from any text before it leaves your hands and reaches an AI chatbot. It works in two layers. The first is an instant regex pass that needs no model and zero download: it finds email addresses, phone numbers, credit-card numbers (validated with the Luhn checksum to cut false positives), IPv4 addresses, US Social-Security-style numbers, and long digit sequences such as account or order IDs. The second is an optional, on-demand layer that loads a small BERT named-entity-recognition model directly in your browser using Hugging Face Transformers.js (the library that lets you run Transformers directly in your browser, with no need for a server) to catch people, organizations, and locations that no regex can reliably find. Inference runs on the ONNX Runtime via WebAssembly (WASM), with optional WebGPU acceleration on supported devices. Every detected value is replaced with a stable, labeled token like [EMAIL_1] or [PERSON_2], and the same value always maps to the same token so the text stays coherent. You see the original and redacted text side by side, a count of what was removed by type, and a full token map you can keep private to map the AI answer back to real values. Everything runs locally in your browser. Your text is never uploaded to any server.
More Free Tools
More than 20 free AI tools.
Why Redact PII Before Pasting Into ChatGPT or Claude
AI chatbots are incredibly useful for drafting replies, summarizing documents, and debugging code, but every message you send to a cloud AI service leaves your device. If that message contains customer emails, phone numbers, card numbers, or someone's name, you have just shared regulated personal data with a third party. For many teams that is a policy violation, and under privacy regimes like GDPR it can carry real consequences.
This PII Redactor solves the problem at the source. Instead of trusting that a provider will handle your data correctly, you remove the sensitive parts before they ever leave your browser. You keep the full context and structure of your text, so the AI still gives you a useful answer, but the actual private values are replaced with neutral placeholders that mean nothing on their own.
How The Two-Layer Detection Works
The first layer is a set of carefully tuned regular expressions that run the instant you type. It recognizes the structural fingerprints of personal data: the shape of an email address, the digit grouping of a phone number, the length and Luhn checksum of a credit-card number, the four octets of an IPv4 address, the format of a Social-Security-style number, and long runs of digits that usually represent account or order IDs. Because it is pure pattern matching, it needs no model and no network, and it works offline.
The second layer is optional and on-demand. Names, companies, and places do not follow a fixed pattern, so catching them requires understanding language. When you enable it, the tool loads a compact BERT named-entity-recognition model (Xenova/bert-base-NER) with Hugging Face Transformers.js, which runs Transformers directly in your browser using the ONNX Runtime on WebAssembly, with optional WebGPU acceleration. It reads your text locally and flags people, organizations, and locations, which then get redacted alongside the regex matches.
Both layers feed into one consistent token map. Each unique value gets a stable label such as [PHONE_1] or [PERSON_2], and repeated values reuse the same token, so the redacted text reads naturally and you can confidently reverse the process later.
Built On Hugging Face Transformers.js
The name-detection layer is powered by Transformers.js, the Hugging Face library whose mission is state-of-the-art machine learning for the web. Its core idea is simple and powerful: run Transformers directly in your browser, with no need for a server. It is designed to be functionally equivalent to the popular Python transformers library, so the same families of models, including BERT, DistilBERT, and RoBERTa, can run client-side instead of on a backend.
Under the hood, models are converted to the ONNX format and executed with the ONNX Runtime compiled to WebAssembly (WASM), with optional WebGPU acceleration for a hardware-accelerated speedup on supported devices. Quantized weights (data types such as q8 and q4) keep the download small and inference fast. For this tool we use Xenova/bert-base-NER, a BERT model fine-tuned for token classification, to perform named-entity recognition on people, organizations, and locations entirely on your machine.
Because Transformers.js is open source under the Apache 2.0 license, the entire detection pipeline is transparent and auditable. There is no proprietary cloud endpoint deciding what counts as personal data, and no hidden call sending your text away for processing. You get a genuine state-of-the-art NER model with the privacy guarantees of local-only execution.
Private By Design, Useful In Practice
Privacy tools only work if you actually trust them, which is why this one does all of its work where you can see it: inside your own browser tab. There is no server to send data to, no account to create, and no logging. You can verify this yourself by watching your network tab; once the page (and optionally the model) has loaded, your text never goes anywhere.
In daily use that means you can paste a real support ticket, a noisy server log, a contract clause, or a spreadsheet row, get a clean redacted version in seconds, copy or download it, and hand it to ChatGPT, Claude, or Gemini without exposing the people behind the data. When the AI responds, your private token map lets you restore the original values locally. It is a small, fast habit that meaningfully shrinks your data-exposure surface every time you use AI.
How It Works
Paste or type any text containing personal data into the input box. An instant regex layer immediately flags emails, phone numbers, credit cards, IP addresses, SSN-like numbers, and long ID numbers without loading anything.
Optionally turn on Detect names to load a small named-entity model in your browser that also redacts people, organizations, and locations. The model downloads once and is cached for next time.
Review the side-by-side original and redacted views, check the count and token map, then copy or download the redacted text and paste it safely into ChatGPT, Claude, or any other AI chat.
Need expert help with AI?
Looking for a specialist to help integrate, optimize, or consult on AI systems? Book a one-on-one technical consultation with an experienced AI consultant to get tailored advice.
Key Features
Privacy & Trust
Use Cases
Limitations
- Regex detection is pattern-based, so it can miss unusual formats and may occasionally over-match number-heavy text. Always review the redacted output before pasting.
- Name, organization, and location detection depends on the BERT NER model, which is not perfect; it can miss entities or mislabel them, especially for non-English or rare names.
- The BERT NER model (Xenova/bert-base-NER) downloads roughly 110 MB on first use and needs a reasonably modern browser with WebAssembly support; the regex layer has no such requirement.
- WebGPU acceleration is used only where the browser supports it; on devices that fall back to plain WASM, the first inference on long text can take a few extra seconds.
- This tool reduces exposure but is not a legal guarantee of compliance. Treat it as a strong first line of defense, not a substitute for your own review.
Q&A SESSION
Got a quick technical question?
Skip the back-and-forth. Get a direct answer from an experienced engineer.
Frequently Asked Questions
Is the PII redactor free?
Yes, it is completely free with no signup, no account, no API key, and no usage limits. It is built on Hugging Face Transformers.js, which is open source under the Apache 2.0 license, so there is nothing to pay for and no paywall on any feature.
Is my text sent to a server or uploaded anywhere?
No. All redaction happens locally in your browser. The instant regex layer never touches the network, and even the optional BERT name-detection model runs on your own device after a one-time download, because Transformers.js runs the model directly in your browser with no need for a server. Your text is never transmitted, never logged, and never stored. You can confirm this in the Network tab of your browser DevTools.
Do I need to install anything?
No. It runs entirely in your web browser, so there is no app, extension, or library to install. The optional name detection downloads a model file the first time you enable it, but that happens automatically in the page and is cached by your browser for next time.
What kinds of personal data does it detect?
The instant regex layer detects email addresses, phone numbers, credit-card numbers (checked with the Luhn algorithm to reduce false positives), IPv4 addresses, US Social-Security-style numbers, and long digit sequences such as account, invoice, or order IDs. With the optional BERT NER model enabled it also detects people names, organizations, and locations through named-entity recognition.
What are the tokens like [EMAIL_1] for?
Each piece of personal data is replaced by a labeled placeholder so the AI still understands the structure of your text without seeing the real value. The mapping is consistent, so the same email always becomes the same token. You can keep the token map and use it to swap the real values back into the AI response after you get an answer.
Do I need to load the model to use the tool?
No. The regex layer works instantly with zero downloads and covers the most common high-risk data such as emails, cards, phones, and IPs. The BERT NER model is only needed when you also want to redact names of people, companies, and places, and you opt into it with a single toggle.
Why is the first load slow, and how big is the model?
The first time you enable name detection, the browser downloads the BERT NER model (Xenova/bert-base-NER, roughly 110 MB) and Transformers.js compiles the ONNX Runtime WebAssembly engine. That one-time setup is what makes the first run slower. After that, your browser caches the model and runtime, so later sessions start quickly, and on supported devices WebGPU speeds up inference further.
Does it work offline or on mobile?
The regex layer works fully offline once the page has loaded, with no network access at all. The name-detection model needs to be downloaded once while online, after which it is cached and can run offline too. It works on modern mobile browsers that support WebAssembly, though large texts run faster on a desktop with WebGPU.
Is this safe to use with confidential or regulated data?
Because nothing leaves your browser, it is far safer than pasting raw data into a cloud AI service. It helps you follow internal data-handling and GDPR-style rules. That said, automated detection is not flawless, so always review the redacted text before sharing it. Treat the tool as a strong safeguard, not a legal guarantee.
How do I get the real values back into the AI answer?
Copy the token map before you paste. After the AI replies using the placeholders, find each token in the answer and replace it with the original value from the map. Because the mapping is one-to-one and consistent, this restoration is straightforward.