Privacy-first governed AI
Your AI conversations, governed by you.
Before your AI answers — or touches anything — SafeChat asks: is this okay? Private info stays in. Interference stays out. Every conversation stays yours.
Before your AI answers — or touches anything — SafeChat asks: is this okay? Private info stays in. Interference stays out. Every conversation stays yours.
All three tiers are governed by the same privacy-first proxy. Pick the one that fits how you use AI.
A lightweight Chrome extension that scans your AI inputs on-device before they ever leave your computer. It detects PII, credit card numbers, API keys, and prompt-injection attempts in real time — and warns you before you hit send. No data ever leaves your machine. Works with ChatGPT, Claude, Gemini, and any AI chat you use in the browser.
A full governed AI chat experience with built-in safety at every step. Every message is scanned for PII and injection attacks before the model sees it, logged to a tamper-evident audit trail, and covered by our no-training pledge. Includes the browser extension. Ideal for professionals who chat with AI daily and need the same protection their enterprise would demand — without the enterprise price tag.
AI safety designed for children and families. SafeChat Kids filters inappropriate content, blocks data sharing, and gives parents full visibility into every AI conversation. No adult content, no data collection, no hidden fees. Built on the same governance engine that protects enterprises — adapted for the people who need it most.
| Feature | SafeChat Free | SafeChat Pro | SafeChat Kids |
|---|---|---|---|
| On-device input scanning | |||
| PII & credit card detection | |||
| Prompt-injection blocking | |||
| Works on any AI chat site | |||
| Governed AI conversations | — | ||
| Tamper-evident audit log | — | ||
| No training on your data | |||
| Age-appropriate content filter | ✓ | ||
| Parent dashboard & history | ✓ | ||
| Custom policy engine | — | — | — |
| Dedicated proxy instance | — | — | — |
| Enterprise SSO & RBAC | — | — | — |
| SOC 2 / HIPAA-ready | — | — | — |
| Priority support & training | — | — | — |
A Chrome extension that sits between you and any AI chat website (ChatGPT, Claude, Gemini, etc.). It scans everything you type before it leaves your device.
When you type or paste into an AI chat box, the extension checks the text on-device for sensitive patterns — email addresses, phone numbers, credit card numbers (using the Luhn algorithm), API keys, SSNs, and known prompt-injection phrases like "ignore all instructions." If it finds something, it warns you with a color-coded badge: green for clean, yellow for mild risk, red for blocked.
No data ever leaves your machine during the scan. Only scan metadata (pattern type, risk level — no raw text) goes to the backend for audit logging. The scan runs on keystroke (debounced), on paste (immediate), and on send (final check). Built on Manifest V3 with a pure JS scan engine, open source under MIT-0.
No account needed — install from the Chrome Web Store and it just works.
A full hosted AI chat experience with governance baked into every message. You talk to the model through the NIBVOK security proxy, which inspects, blocks, and logs every interaction before the AI sees it.
Every prompt passes through the NIBVOK runtime gateway — the same engine that protects enterprise agent deployments. PII and injection patterns are blocked at the proxy layer, not just warned about. Each allow/confirm/deny decision is written to a tamper-evident hash chain (SHA-256, each entry links to the previous via prev_hash). You can download the log and verify it yourself with verify-audit-chain.mjs.
Key difference from the free extension: The extension is client-side only — it warns you but doesn't stop the message if you override. Pro enforces the block server-side. If the proxy detects a credit card or injection attempt, the message never reaches the model. Period. Also includes the browser extension at no extra cost.
Ideal for professionals who chat with AI daily and want enterprise-grade data-loss protection without the enterprise contract.
The same governance engine, tuned for children and families. Filters out adult content, violent language, self-harm references, and inappropriate topics before they ever reach the AI — and before the AI generates anything harmful back.
The proxy uses an additional content-classification layer tuned for child safety. Inappropriate prompts are blocked with a kid-friendly message ("Let's ask something else together!"). The parent dashboard shows the full conversation history — every prompt, every response — so you know exactly what your child is asking AI and what it is telling them.
No data is collected, no profiles are built, no advertising. No login required for the child (they use a family-share code through the extension). The model uses a safety-tuned system prompt that gently redirects inappropriate queries rather than answering them.
Recommended for ages 8–15 with parent supervision. Built on the same governance engine that protects enterprises — adapted for the people who need it most.
Every message passes through the NIBVOK Security Gateway before the model sees it. On every tier.
Each prompt is scanned for PII, credit card numbers, secret keys, and prompt-injection patterns before it reaches inference.
Risky content never reaches the model. You get a clear reason — not a black-box denial.
Every Allow / Confirm / Deny decision is written to a tamper-evident chain. Yours to review, yours to keep. Stated in R-18.
From free browser protection to full enterprise governance. No credit card needed to start.