Join the waitlist
Early access

Your secrets stay with you. And we can't see them.

Sight4 catches sensitive information, like keys, card numbers and passwords, before it's sent to AI, right on your device. Your prompts never come to Sight4, and we never decrypt your traffic.

Join the waitlist See how it works
What a stop looks like on an AI chat site. Example with fake, shortened data: the prompt contained an API key, so Sight4 stopped it before it was sent.

How Sight4 protects you

Sight4 checks what you send to AI right on your device. It stops API keys, card numbers and Social Security numbers before they're sent, and asks you about passwords.

Here's what works today and what we're building next. We'd rather tell you what isn't done yet than oversell it.

Now Working today, in early access

In your browser

ChromeEdge: soonFirefox: soon

The Sight4 extension checks prompts on ChatGPT, Gemini, Grok and Kimi before they're sent, with more AI chat sites, including Claude, being tested now. It looks at what you type, paste or drop in, and, if you turn on file checks, the files you attach.

It stops API keys, card numbers and Social Security numbers, and asks you to check tokens and passwords before sending. It tells you what kind of data it found and how to fix it, never the data itself.

The check runs in your browser. Nothing is sent to Sight4.

In AI coding tools

Claude CodeCursorGemini CLI: planned

Sight4's desktop helper plugs into the tools' own hooks. It checks each prompt before the tool sends it, lets you choose in a settings file which MCP servers the AI may use, and checks what those tools send back before the AI reads it.

It's a command-line tool for now. A simple installer is on the way.

Coming soon On the roadmap

The Sight4 portal

One place to see which devices are covered and set the same protection for your household or team. It will handle settings and counts only, never what you wrote.

Tricks typed into the chat

Direct prompt injection

Spotting prompts written to talk an AI out of its own rules, often called jailbreaks.

Hidden instructions in AI replies

Indirect prompt injection

Checking replies in AI chats for instructions planted there. In coding tools, Sight4 already checks what connected MCP tools send back.

A checkup for AI rule files

Flagging hidden instructions in the files coding tools follow automatically, such as .cursorrules or CLAUDE.md.

Phones and tablets

Bringing the same checks to Safari on iPhone and iPad.

What's prompt injection?

It's when someone slips instructions to an AI that you didn't write, to make it do something you didn't ask for, like revealing your information or ignoring its safety rules.

Direct

The instructions are in what's typed or pasted into the chat. For example, a "magic prompt" copied from a forum that tells the AI to ignore its rules.

Indirect

The instructions are hidden in something the AI reads for you: a web page, a document, an email, a reply, or a tool it's connected to. You may never see them, but the AI might follow them.

(No, actually)

We can't see your prompts, by design.

Most privacy promises are about what a company chooses not to look at. Ours is about what Sight4 is built to be unable to do. "We" means Sight4: the AI site you're chatting with still gets whatever you choose to send, just as it would without us.

Now

The extension is built not to call home

Its pages block every connection (connect-src 'none'), and the code that reads your prompts is checked at every build, which fails if network code gets in. It asks your browser only for local settings and the AI sites it checks, plus Google Search if you turn on the optional switch that hides Google's AI answers. The detection rules are built in, so nothing is downloaded either.

Now

The check happens on your device

Your prompt is checked in your browser, or on your computer, before it's sent. Sight4 keeps only counts and the kind of thing it found, like "API key", never the text.

Now

We never decrypt your traffic

No certificate to install, and nothing in your internet connection. Sight4 works through the browser and the AI tools' own hooks instead.

Now

The desktop helper uploads nothing

The helper for AI coding tools keeps a short log on your computer of what it decided and what kind of data it found. Nothing is uploaded, and your prompt text is never written down.

See exactly what leaves your device

For developers and vibe coders

Your AI coding tool can read your whole project, including the .env you forgot about. Sight4 plugs into the hooks Claude Code and Cursor already have, so it checks each prompt and each MCP tool call before it goes out. Files the tool opens on its own aren't checked yet.

  • Keys stay out of prompts

    API keys, access tokens and private keys are stopped before the tool sends the prompt, with a note on what to remove. Passwords and web tokens (JWTs) get a warning.

  • You pick the MCP servers

    List the servers you trust in Sight4's settings file, and Sight4 stops the rest. On the Strict setting, it asks you before any server you haven't listed.

  • Tool results get checked too

    Before the AI reads what an MCP tool sent back, Sight4 looks for planted instructions and secrets. In Claude Code it holds back results with strong signs of planted instructions and blanks out keys found in meeting notes; other findings show as warnings. On Cursor everything shows as a warning for now.

  • No root certificate, no traffic interception

    Sight4 doesn't sit in your network connection to the model provider. Your keys and your traffic stay yours.

Early access is command-line only. Join the waitlist and pick "Developer" to try it.

What leaves your device

Nothing is sent to Sight4.

This is the list behind "We can't see your prompts." The extension and the desktop helper check everything on your device. Today there's no account and no telemetry, and the extension makes no network connections of its own.

Sent to Sight4

Nothing.

Stays on your device

  • Your prompts, pastes and drops, and the files you attach when file checks are on. Sight4 checks them where they are.
  • What was found. Sight4 keeps only counts and the kind of data, such as "API key", never the text itself.
  • Your settings and the detection rules, which are built in, so nothing is downloaded.

Still goes to the AI site

  • Prompts and files you choose to send, as they would without Sight4.
  • What you type in the site's prompt box. The site can read it before you press send, even when Sight4 stops the send.
  • The site can tell that a Sight4 notice appeared, but not what it says. It can also see what you clicked, and which pasted text Sight4 blurred.

Planned: the Sight4 portal for households and teams. When it arrives, this list changes first, before anything new is sent. What's sent then will be metadata only, like counts and kinds of data, never what you wrote. The details are in our privacy policy.

Who it's for

Home and family

You use AI to write, plan and fix things. Sight4 catches the card number you didn't mean to paste, and asks before a password goes out. Household settings to manage everyone's computers together are on the roadmap.

Small businesses and MSPs

Keep card numbers, Social Security numbers and company keys out of AI chats without needing IT staff. Team settings for businesses and the people who look after their computers are on the roadmap.

Developers and vibe coders

Ship with Claude Code or Cursor without pasting your keys into a prompt. Sight4 checks prompts and MCP tool results right where the tool runs.

Join the waitlist

Sight4 is in early access. Tell us who you are and we'll let you know when it's ready for you.

I'm mostly interested for

This opens your email app with a short note to info@sight4.com. There's no sign-up service: your message goes to our mailbox, and we'll use your address only to tell you about Sight4. See our privacy policy.