Pre-Deployment Security Platform

Validate your AI
before it ships.

Sentryɸ runs multi-layer security analysis, adaptive vulnerability detection, and model behavior stress testing — in a single scan. One report. Every risk. Actionable fixes.

Free & open source  ·  Windows 10+  ·  Ubuntu 20.04+  ·  No account required

Multi-Layer Security Analysis· Adaptive Vulnerability Detection· Model Behavior Stress Testing· Pre-Deployment Validation Pipeline
The Problem

LLMs ship with
unknown risks.

Most teams deploy language models with no security validation. The ones that do run a single scanner — missing entire categories of vulnerabilities that only surface under different testing methodologies.

!Fragmented tooling with incompatible outputs
!No unified scoring across analysis methods
!Manual setup blocks adoption at every step
The Solution

One platform.
Complete coverage.

Sentryɸ orchestrates multiple security analysis layers into a single, automated pipeline. Input your model endpoint. Get a unified, scored, and actionable security report — in minutes, not days.

✓Normalized vulnerability schema with evidence
✓Dual-axis scoring: severity + confidence
✓Three levels of remediation per finding

Everything you need to validate an LLM.

Not a single scanner. A complete pre-deployment security platform.

◎

Multi-Layer Scanning

Multiple security analysis engines run in parallel. No duplicate setup. No format translation. Full cross-method finding corroboration.

OWASP LLM TOP 10
⊞

Explainable Scoring

Severity and confidence are scored independently — with a plain-English rationale for every number. No black-box risk ratings.

DUAL-AXIS SCORING
⬡

Secure Deployment

The Deployment Advisor reads your risk profile and configures guardrails, IAM policies, and monitoring alarms. One click to apply.

CLOUD DEPLOYMENT
⟐

Actionable Remediation

Every finding includes fixes at model level, system level, and deployment level. A vulnerability report that tells you how to fix it.

THREE-LAYER FIX
⊡

Multi-Format Reports

PDF for stakeholders and compliance. HTML for sharing. JSON for CI/CD pipelines. Every format includes the OWASP coverage map.

PDF · HTML · JSON
⊘

Privacy by Architecture

API keys in OS keychain only. Probe content encrypted at rest. Local model scans are fully air-gapped. Zero telemetry by default.

ZERO TRUST

From model to secure deployment.

Four steps. No setup. No terminal. No guesswork.

01

Connect

Paste your API endpoint and key — or drop a local model file. Sentryɸ validates the connection and estimates cost before anything runs.

02

Configure

Choose Quick, Standard, or Deep scan depth. Review estimated cost, duration, and test categories. Nothing executes without your explicit approval.

03

Scan

Multiple analysis layers run in parallel. Findings appear in real-time. The rate limiter protects your API budget. Pause or cancel anytime.

04

Ship

Your unified report is scored, explained, and actionable. Every finding has remediation. Deploy with confidence using generated cloud configurations.

Three analysis layers.
One unified report.

Each layer tests a different failure mode. Running only one gives you partial coverage.

LAYER 1

Probe-Based Analysis

Systematic, reproducible probing with 100+ test patterns. OWASP-mapped. Deterministic results across runs.

Detects: prompt injection, jailbreak bypasses, refusal failures
LAYER 2

Adversarial Simulation

Multi-turn adversarial dialogue simulation. Tests conversation-dependent vulnerabilities that only surface across extended interactions.

Detects: session escalation, context manipulation, persona drift
LAYER 3

Behavioral Evaluation

Metric-driven evaluation framework. Quantitative thresholds for bias, hallucination, toxicity, and factual consistency.

Detects: systematic bias, factual drift, toxicity patterns

From report to secure infrastructure.

The Deployment Advisor reads your risk profile and generates production-ready cloud configurations. For AWS: one-click execution.

CapabilityAWSAzureGCP
Estimated monthly cost$142/mo$167/mo$139/mo
Native guardrails✓ BedrockManual—
One-click deploy✓——
IAM auto-configuration✓——
Monitoring alarms✓ CloudWatchManualManual

See it in action.

Real-time scan progress with category tracking and live finding detection.

Sentryɸ — Security Scan
Input
Config
Scanning
Results
Targethttps://api.example.com/v1/chat
Executing security probes73%
73 / 100 probesElapsed: 47s
Category Progress
Prompt InjectionComplete
Jailbreak ResistanceComplete
Data Leakage DetectionActive
Behavioral AnalysisQueued
Live Findings
System prompt extraction via multi-turn dialogueCritical
Jailbreak via role-play instruction injectionHigh
Inconsistent refusal boundary for medical queriesMedium

OWASP LLM Top 10 coverage.

We show what we cover — and what we do not.

LLM01Prompt InjectionFull
LLM02Insecure OutputFull
LLM03Training Data PoisonPartial
LLM04Denial of ServicePartial
LLM05Supply ChainAdvisory
LLM06Sensitive InformationFull
LLM07Insecure Plugin DesignN/A
LLM08Excessive AgencyPartial
LLM09OverrelianceFull
LLM10Model TheftNot in scope

Incomplete coverage maps are a red flag — we don't hide ours.

What comes next.

Active development. Public roadmap. No guessing.

v1.0 — Foundation

Released October 2026
Multi-layer security analysis orchestration
Unified VulnerabilityFinding schema v1.0
Explainable dual-axis confidence scoring
AWS one-click deployment with guardrail configuration
Report export: PDF, HTML, JSON

v1.5 — Developer Workflow

Target: Q1 2027
CLI interface for terminal-first developers
GitHub Actions integration for CI/CD pipelines
Azure + GCP one-click deployment
Scan comparison and regression detection

v2.0 — Platform

Target: Q3 2027
Team collaboration and shared dashboards
Native GitHub App for automated scanning
Model version tracking and regression history
Multi-modal scanning for vision-language models

Stop guessing.
Start scanning.

Free. Open source. No account required. No server.
Download and run your first scan today.

v1.0.0  ·  Windows  ·  Linux  ·  MIT License