SENTINEL RED

What SENTINEL RED is

SENTINEL RED is a free, open-source suite for security and quality testing of applications built with large language models. Six modules address prompt injection, hallucinations, data leakage, adversarial testing, poisoning, and compliance, with more than 85 attack patterns listed. Tests can target APIs or local systems, hosted models or local LLMs, using user-provided adapters. A typical run involves adding a model target and API key, choosing modules and test depth, watching live results in a WebSocket dashboard, and generating a PDF report with scoring. The REST API can create runs, poll their status, and download reports programmatically. SENTINEL can run locally, in pipelines, or on schedules. The recommended deployment uses Docker Compose and opens the dashboard at localhost:3000. OpenAI and Anthropic keys are shown as configuration examples, with multiple providers supported through environment variables or dashboard settings. Its local-first design is intended to keep secrets on the user’s machine. The project is self-hosted and licensed under AGPL-3.0.