SENTINEL — LLM Trust & Safety for VS Code

Real-time LLM prompt security scanning directly in VS Code.

Screen any text for prompt injection, PII leakage, jailbreak attacks, toxicity, hallucination, and 14+ other threat categories — powered by SENTINEL's 19-agent parallel security mesh.

Features

🔍 Screen Selected Text

Right-click any selected text and choose "SENTINEL: Screen Selected Text" to instantly scan it through all 19 security agents.

📊 Trust Score

Get a 0-100 trust score for any highlighted text. Scores below 70 indicate potential threats.

📄 Full File Scan

Scan all string literals in the current file for potential prompt injection or PII leakage.

Gateway Status — Live connection health, agent count, FAISS vector count
Scan Results — History of all scans with decision/score/latency
Agent Mesh — All 19 agents organized by version tier

🎨 Inline Decorations

After scanning, threat scores appear inline next to the scanned text:

🔴 ⛔ BLOCK 85% — High-risk content blocked
🟡 ⚠️ WARN 45% — Moderate risk, content sanitized
🟢 ✅ SAFE 12% — Content is safe

⚡ Auto-Scan on Save

Enable in settings to automatically scan all string literals when you save a file.

19-Agent Security Mesh

Tier	Agents	Capabilities
v1 Core	InjectionScout, PIISentinel, JailbreakGuard, ToxicityScreener, HallucinationProbe, ContextAnchor, ComplianceTagger	FAISS ANN, SpaCy NER, Detoxify, NLI grounding
v2 Enterprise	ResponseSafety, MultilingualGuard, LocaleComplianceRouter, ToolCallSafety, BrandGuard, TokenAnomaly	Universal model support, SSRF detection
v3 Intelligence	PromptLineage, IntentClassifier, AdversarialRephrasing	13-label intent, semantic mutation detection
v4 Advanced	JailbreakPatternDetector, CostAnomalyDetector, AgenticLoopBreaker	65+ signatures, DAN attacks, loop detection

Configuration

Setting	Default	Description
`sentinel.gatewayUrl`	`http://localhost:8000`	SENTINEL Gateway URL
`sentinel.tenantId`	`""`	Your tenant ID
`sentinel.apiKey`	`""`	API key (Pro tier)
`sentinel.autoScan`	`false`	Auto-scan on save
`sentinel.showInlineScores`	`true`	Show inline decorations
`sentinel.blockThreshold`	`0.7`	Block threshold (0-1)
`sentinel.warnThreshold`	`0.35`	Warning threshold (0-1)

Sentinel LLM AI Security

Sentinel Labs

SENTINEL — LLM Trust & Safety for VS Code

Features

🔍 Screen Selected Text

📊 Trust Score

📄 Full File Scan

🛡️ Sidebar Dashboard

🎨 Inline Decorations

⚡ Auto-Scan on Save

19-Agent Security Mesh

Configuration

Getting Started

Requirements

Links

License