Sentinel AI

•
22
•

Provide real-time safety guardrails for Large Language Model (LLM) applications with sub-millisecond latency.

Sentinel AI is a lightweight, zero-dependency safety layer designed to protect your LLM applications from a wide range of threats, including prompt injection, PII leaks, harmful content, and code vulnerabilities. It offers comprehensive scanning capabilities covering the OWASP LLM Top 10, delivering results with exceptional speed. The tool is highly versatile, providing Python and TypeScript SDKs, a REST API, and various integrations for popular LLM frameworks and services like Claude, OpenAI, LangChain, and LlamaIndex. It also includes robust features for production environments, such as streaming protection, authentication, rate limiting, and OpenTelemetry.

主要功能

01Achieves ~0.05ms average scan latency without requiring GPUs or external API calls.
02Offers 11 built-in scanners covering the OWASP LLM Top 10, including prompt injection, PII, harmful content, and code vulnerabilities.
03Core library boasts zero heavy dependencies, relying only on `regex` for minimal footprint.
04Provides drop-in integrations and middlewares for Claude, OpenAI, LangChain, LlamaIndex, and FastAPI.
05Supports multiple deployment modes including Python/TypeScript SDKs, REST API server, LLM API firewall, and MCP server.
060 GitHub stars
07Includes comprehensive CLI tools for project-wide security scanning, auditing, and pre-commit hooks.

使用案例

01Validate MCP tool definitions and instruction files like CLAUDE.md for hidden injection vectors and security vulnerabilities.
02Safeguard LLM applications deployed via FastAPI or as a transparent proxy for Anthropic/OpenAI APIs.
03Prevent prompt injection and PII leaks in LLM interactions by scanning inputs and outputs in real-time.
04Integrate robust security checks into CI/CD pipelines using project-wide scans, pre-commit hooks, and Claude Code hooks.