nGuard
Sign inOpen dashboard
Security testing across applications and AI systems

Find what blocks a product from being ready to ship.

nGuard reviews repositories, live URLs and agent endpoints, then returns prioritised findings with proof, a readiness score and hardening guidance.

Start a project Open the dashboard
An AI model, tools and data in a row, with a red approval boundary between the model and its tools and a skipped approval flagged
What nGuard covers

Code, runtime and agents, in one readiness score.

Connect a repository, a deployed URL, an LLM endpoint, or all three. Each surface gets its own score; the project carries the overall.

A getAccount(id) handler beside two account folders with a missing ownership check
Code

Secrets, dependencies and access paths.

Static analysis · NRS
Request, session scope and response, with the session scope outlined in red
Runtime

APIs, sessions and deployed behaviour.

Live-URL scan · NQS
Plan, approval and action blocks, with the approval check outlined in red
Agents

Instructions, tools and action boundaries.

Red team · Agent score
How a run works

From probe to proof.

The deeper engine detail lives inside the run report. The path from target to hardening plan is the same every time.

01

Profile the target

nGuard detects the stack, routes, services, auth patterns, live surface, and AI-specific behaviour where present. You do not describe the system first.

02

Run the analysis

Static, runtime and AI checks run with progress per engine. Partial engine failures are reported instead of hiding uncertainty, and unavailable checks are marked rather than pretended.

03

Return the scorecard

A readiness score, prioritised findings, machine evidence for each one, and the next actions required before launch.

04

Fix and verify

Every finding carries fix guidance and a generated test for your pipeline. Re-run the scan to prove the boundary now holds.

Coverage

Standards and scan families, visible upfront.

Security teams should not need a demo to understand the surface area. Findings are tagged against the frameworks security, audit and AI-risk teams already recognise.

Mapped standards

OWASP Top 10OWASP API Security Top 10OWASP LLM Top 10CWECVESOC 2HIPAAGDPRPCI-DSS

Scan families

  • Secrets and exposed credentials
  • SAST and insecure code paths
  • Dependency and supply-chain CVEs
  • Auth, JWT, session and cookie issues
  • Headers, CORS, CSP, HSTS and rate limits
  • Deployment, Docker, CI/CD and health checks
  • Reliability, resilience and observability
  • Data integrity, API quality and audit trails
  • Live URL DAST and passive runtime checks
An adversarial test environment: prompt injection, data leakage, tool misuse, authorisation and runtime probes wired into a model, tools and data, with one finding traced to its evidence and fix

Red team & DeepTeam

Multi-turn probes, injection and unsafe actions.

AI-specific checks are detected, run automatically and included in the scorecard. A finding is grounded in the tool trace and the missing check, not the wording alone.

  • Prompt injection and jailbreak attempts
  • System-prompt and context leakage
  • RAG and knowledge-base exposure
  • Tool misuse, goal hijacking and unsafe actions
  • MCP, agent configuration and permission risks
  • Cost, iteration and human-approval controls
Evidence

Every score change has a reason you can read.

The report is the product moment. It explains risk, proof and next action without asking anyone to trust a black-box number.

  • File and line references for code findings
  • Request and response proof for live findings
  • Severity, category, CWE/CVE and compliance mapping
  • Fix guidance, generated tests, and exportable reports
HighApproval bypassAction-taking agent · refund tool
The approval boundary is not enforced at the tool.
Tool trace · POST /tools/refund
POST /tools/refund
{ amount: 120, approval_ref: null }
Response → action accepted
Fix

Validate approval server-side, bound to the user, action and amount. Reject missing or mismatched approval, then re-run the scenario.

Ready to run

Start with the asset you have.

Connect source code for full analysis, or scan a deployed URL first when repository access is not available.

Start a project Sign in