Argus Lens — AI Prompt Guard

Scan AI prompts for secrets, PII, and prompt injection before you send them — locally, on ChatGPT, Claude, and Gemini.

You paste code, logs, and drafts into AI chats every day. One paste can leak an AWS key, a customer's email, or your own password — and once it's sent, you can't unsend it. Argus Lens is a guardrail that sits in the chat page and checks your prompt before you send it: • Secrets → BLOCKED. AWS access keys, GitHub / OpenAI / Anthropic API keys, private keys, Slack tokens, JWTs, and password=/api_key= patterns stop the Enter key and the Send button. • PII → ONE-CLICK REDACT. Emails, phone numbers, SSNs, credit card numbers, and IPs are flagged with a Redact button that rewrites your draft in place. • Prompt injection → FLAGGED. "Ignore all previous instructions", role-override, and system-prompt probing patterns are called out before they go anywhere. • Your own blocklist. Add project codenames or client names as patterns in Options — matches are blocked. • Three policy profiles — Strict, Standard, Permissive — ported from the open-source Argus gateway's policy file. • A local audit trail. Blocked and redacted events are logged with redacted matches only (AKIA••••), stored only in your browser, clearable any time. 100% local. No server, no analytics, no account, no data ever leaves your browser. Argus Lens is the browser companion to Argus, the open-source security layer for AI models: https://github.com/karmendra8386/argus — the detectors and policies here are the same ones the gateway enforces for API calls. Works on chatgpt.com, claude.ai, gemini.google.com, and Google AI Studio.