Redact
Show HN: On-device Chrome extension that blocks credential leaks to LLM chats
Last verified:
What is Redact?
Redact is a Chrome extension that protects people from accidentally pasting secrets into AI chat apps. It watches paste events in supported LLM chats and scans the pasted text locally in the browser before anything is sent to the provider. It is positioned as a privacy guard for tools like ChatGPT, Claude, Copilot, Perplexity, Grok, DeepSeek, Mistral, and Poe.
The core idea is to catch credentials and personally identifiable information before they leave the user’s browser. According to the site, detection runs on-device with a local neural network and completes in about 150 milliseconds, so the user can still paste quickly while getting a warning or block when sensitive content is detected. The workflow is simple: detect, decide, then paste.
Redact gives users per-data-type control from a popup, with options to block, warn, send a redacted version, send the original, or cancel the paste. The site says the defaults are tuned for engineers, which suggests it is aimed at technical users who frequently copy API keys, tokens, or other sensitive snippets into chat tools.
The product emphasizes privacy as a feature. The website states that paste content stays in the browser, the extension makes no network calls, there is no telemetry, and the model is bundled inside the extension package. It also says the source is available on GitHub under the PolyForm Noncommercial license, making it appealing to users who want auditable, client-side protection rather than server-side scanning.
Redact pricing
Pricing model: Freemium
The website says Redact is free in the Chrome Web Store and that the source is available on GitHub. No paid plans, subscription tiers, usage limits, or enterprise pricing are listed on the site, and no separate free tier is described beyond the product being free to install. Included features shown on the website are browser-based detection, local inference, support for major LLM chat sites, and per-data-type block/warn controls.
Redact pros
- Catches secrets before they reach AI chats
- Runs entirely in the browser
- Local neural network detection
- About 150 ms inference time
- No network calls during scanning
- No telemetry or analytics SDKs
- Supports ChatGPT
- Supports Claude
- Supports Microsoft Copilot
- Supports Perplexity
- Supports Grok
- Supports DeepSeek
- Supports Mistral Le Chat
- Supports Poe
- Per-data-type block or warn controls
- Can send a redacted version instead of raw text
- Can cancel a paste before it is submitted
- Defaults tuned for engineers
- Source available on GitHub
- Model bundled with the extension
Redact cons
- Only works on supported chat sites
- Not a general-purpose system-wide redactor
- Chrome extension only
- Requires installation before protection starts
- Limited to paste-based leakage, not all input methods
- Redaction quality depends on local model detection
- No built-in enterprise admin controls mentioned
- Noncommercial license may limit some users
Frequently asked questions about Redact
What does Redact actually do?
Redact intercepts pasted text in supported AI chat sites and checks it locally for credentials and personally identifiable information before the content reaches the chat provider. If sensitive data is detected, it can warn, block, redact, or let the user cancel the paste.
Which chat apps does Redact support?
The website lists ChatGPT, Claude, Microsoft Copilot, Perplexity, Grok, DeepSeek, Mistral Le Chat, and Poe. It also says the extension is active only on these supported chat sites, with other sites requiring an issue or request on GitHub.
Does Redact send my pasted data to a server?
No. The site says detection runs locally in the browser, the model runs in a Web Worker, and the extension makes no network calls. It also states that there is no telemetry and no analytics or error-reporting SDK.
How does the detection work?
Redact uses a local neural network to scan each paste for credentials and PII. The website says inference takes about 150 milliseconds and runs inside the browser, so the check happens before the paste is delivered to the chat site.
Can I choose whether to block or warn?
Yes. The site says users can set Block or Warn per data type from the popup. It also mentions that the defaults are tuned for engineers, implying users can customize the response based on the kind of sensitive data detected.
Can Redact redact text instead of blocking it?
Yes. The workflow shown on the site includes sending a redacted version, sending the original, or canceling the paste. That means Redact can replace sensitive content rather than only stopping the paste outright.
Is Redact open source?
The website says the source is available on GitHub and describes it as auditable. It also notes that the source is under the PolyForm Noncommercial license.
How fast is it?
The site says detection runs in about 150 milliseconds on a standard laptop. Because the model is local and bundled in the extension, the scan is designed to happen quickly before the paste completes.
How do I install Redact?
The website says to add it to Chrome and describes it as free in the Chrome Web Store. Once installed, it works on the supported chat sites whenever you paste text.
Is Redact free or paid?
The site says Redact is free in the Chrome Web Store. It does not list any paid plans, premium tiers, or subscription pricing on the website.