Frequently Asked Questions

SENTINEL is an AI safety and compliance platform. It works in two ways: a Chrome Extension that detects jailbreak attempts and prompt injection in real time while you browse AI platforms, and a Compliance API that generates tamper-proof certificates proving your AI agents behave within the boundaries you set.

SENTINEL works with 7 AI platforms: ChatGPT, Claude, Gemini, Grok, DeepSeek, Meta AI, and OpenClaw. The Chrome Extension provides real-time browser protection on the first six. The Compliance API works via MCP for Claude and REST API for all other platforms, including any custom AI agent.

Not for the Chrome Extension — install it, paste your API key, and you're protected. For the Compliance API: Claude Code and Claude Desktop integrate with a simple config change (MCP, no code). For other platforms, you integrate via REST API. Our quickstart guide covers both paths.

The extension detects jailbreak attempts and prompt injection patterns using SENTINEL's Behavioral Injection Detection (BID) engine. It scans both your input (what you type) and the AI's output (what it responds) in real time. It uses behavioral analysis — not simple keyword matching — to identify manipulation attempts.

The extension runs SENTINEL's Behavioral Injection Detection engine entirely in your browser. Your conversation text is scored locally — it never leaves your machine. The only network call the extension makes is to validate your API key on setup.

Green means the conversation is safe — no threats detected. Yellow means caution — something scored above the alert threshold but may not be a genuine threat. Red means a jailbreak or injection attempt was detected. Gray means the extension is loading or the BID engine hasn't initialized yet.

SENTINEL sees behavioral measurements you choose to send — things like quality scores, response times, or safety metrics. It does NOT see your AI's prompts, training data, or customer information. The compliance certificates prove your agent is "within limits" without containing the underlying measurements.

Traditional audits happen periodically — quarterly or annually. SENTINEL monitors continuously and links every certificate to the previous one, creating an unbroken compliance timeline. A gap in the chain is immediately visible. You go from "we passed our last audit" to "we've been continuously compliant since January."

Yes. Anyone can verify a SENTINEL certificate without a SENTINEL account or API key. Just share the certificate or a link to our verification page. The mathematics behind the certificate make it self-verifying — no trust in SENTINEL is required.

For Claude Code or Claude Desktop users: about 5 minutes. You add a few lines to your configuration file and you're generating certificates immediately. For custom API integrations, our quickstart guide walks you through the first certificate in about 15 minutes.

We offer a free tier with 100 certificates per month — enough to evaluate and prototype. Our Pro plan is $49/month for 5,000 certificates with priority support. Enterprise plans are custom. Verification is always free and unlimited.

SENTINEL is designed for privacy by construction. The compliance certificates prove behavioral conformance without containing personal data. Since the certificates don't store the underlying observations, there's nothing to delete, anonymize, or subject to data requests.

SENTINEL reports what it measures honestly. If your agent's behavior exceeds your defined limits, the certificate will say "out of bounds." SENTINEL doesn't prevent bad behavior — it proves whether behavior was good or bad. This honest reporting is the foundation of trust.

Yes! Sign up for a free API key, install the Chrome Extension, and you'll see real-time protection on your next AI conversation. Or visit our homepage to see a live demo of compliance certificate generation.