OpenClaw Guardrails

🛡️ OpenClaw Guardrails — Security Toolkit for Autonomous AI Agents
AI agents browse, click, and act on your behalf. OpenClaw Guardrails makes sure they only do what you intended.

⚡ Real-Time Prompt Injection Detection
Pages are scanned as your agent encounters them. Detects instruction overrides, role manipulation, delimiter injection, encoding attacks, and system prompt leak attempts. Visual warnings appear directly on the page so threats are impossible to miss.

🚫 Agent Action Blocking
Automatically hides dangerous action buttons — like sending emails or submitting forms.Your agent can't do what it can't see.

🌐 Custom Site Controls
Whitelist sites you trust. Blocklist sites you don't. Domain-level control over where your agent is allowed to operate freely.

📋 Local Scan History
Every scan is logged locally so you can review what was detected, when, and where. Full audit trail, zero cloud dependency.

🔒 Configurable Severity Thresholds
Decide what level of threat triggers a warning vs. a hard block. Tune the guardrails to match your risk tolerance.

100% local. Zero data leaves your browser.

All scanning runs on-device using the Vard detection library. No page content is ever transmitted externally. We don't collect, store, or share any browsing data. Settings and history stay in local storage, fully under your control.

OpenClaw Guardrails is an independent project and is not affiliated with the opensource project OpenClaw, but is designed to assist users of OpenClaw as well as all other types of autonomous agent.