An AI-driven penetration test you run from your terminal

AI-driven pentest from one command

Point Sentinel at an authorized target and it handles reconnaissance, tool selection and analysis, driving real scanner and exploit tooling rather than a passive scan. You get back a deduplicated, severity-ranked, triaged report instead of a raw tool dump.

{ "target": "https://staging.example.com",
  "workflow": { "default": "web_pentest" } }

Authenticated, multi-tier testing

Sentinel assesses the same target at several privilege levels — unauthenticated, low-privilege, elevated and machine-to-machine — by logging in through standard SSO flows. A cross-tier comparison flags findings that only surface with more access (potential escalation surface) versus those your authentication already blocks.

tiers:
  l0_unauth:  { auth_mode: none }
  l1_session: { auth_mode: cookie }
  l3_bearer:  { auth_mode: bearer }

Verified findings, not scanner noise

Uncertain findings go through a separate adversarial review — multiple models argue for and against each one and a judge flags the false positives — so your team triages higher-confidence issues instead of a wall of alerts. Findings are enriched with CWE/CVE references and CVSS scoring.

Scope hard-guard

Every run is validated against a scope guard — an IP/CIDR deny-list plus an optional allow-list of permitted domains — and refuses any out-of-scope host before it fires a single request. Testing stays on the authorized staging and test targets you declare.

"scope": {
  "allow": ["*.staging.example.com"],
  "deny":  ["*.prod.*"]
}

Standards-based, CI-ready output

Findings export as SARIF 2.1.0 alongside human-readable Markdown and JSON, so results drop straight into existing code-scanning and issue-tracking workflows. Runs support a safe default mode, unattended execution with per-tool prompts skipped, and session resume to continue an interrupted assessment.

"output": {
  "format": "markdown",
  "export": ["sarif"]
}

Pick the assessment per engagement

Choose from shipped workflows — web-app pentest, API pentest, static source review, LLM red-teaming, Active Directory assessment and mobile — and add a new authentication tier or target by editing one declarative config file, with no code change required.

workflow:
  default: web_pentest   # shipped workflow, edit one file