ContextHound 作者: IulianVOStrut

ContextHound scans text you type into AI tools for prompt injection attacks - attempts by malicious content to hijack, manipulate, or jailbreak the AI model you are using.

Scan pill
A lightweight indicator appears next to any AI chat input on any website. As you type, ContextHound analyses the text against 70+ detection rules covering jailbreaks, role-hijack instructions, indirect injection, data exfiltration attempts, system-prompt overrides, and more. A risk score and plain-English explanation appear instantly.

DevTools panel
Open the browser DevTools and select the ContextHound tab to monitor live LLM API traffic. The panel intercepts and scans requests to OpenAI, Anthropic, Google Gemini, Mistral, Groq, Cohere, DeepSeek, and others.

Popup scanner
Click the toolbar icon to paste and scan any text manually before sharing it with a model.

Privacy
Everything runs entirely in your browser. No text, prompt content, scan results, or metadata is collected, stored externally, or transmitted anywhere. The extension has no backend server. Full source code is available on GitHub.