Skip to content

Security is the architecture, not a feature

Every design decision in CitrusGlaze starts with the threat model. 17 Rust crates, 560+ tests, and a 9-stage inspection pipeline. The scanner is open source — audit the detection patterns yourself. Coverage informed by the MITRE ATLAS AI threat landscape.

Threat model

Each threat maps to implementing code. Click the defense to view the source.

Threat categories informed by the MITRE ATLAS framework for AI systems.

ID Threat Defense
T1 Data Exfiltration / Insider Threat citrusglaze-secrets + honey tokens
T2 Supply Chain destination allowlist + Sigstore attestation
T3 Destructive Ops Cedar policies + injection detection
T4 Lateral Movement destination allowlist + metadata endpoint deny
T7 Denial of Service Rate limiting + cost caps
T8 Loop Attack loop_detection.rs — 5 in 60s threshold

Testing

560+

Tests across 17 crates

Unit tests, integration tests, and end-to-end tests covering the full inspection pipeline.

43+

Adversarial bypass tests

Base64 encoding, URL encoding, hex encoding, ROT13, Unicode confusable characters — all tested.

39+

AI tool compatibility

End-to-end compatibility tests for Claude Code, Cursor, Copilot, ChatGPT, and more.

CI

Continuous integration

GitHub Actions CI runs on every commit across macOS and Linux. Clippy warnings treated as errors.

Transparency

Scanner is MIT licensed

Audit the scanner's 349+ detection patterns. No obfuscation, no hidden calls.

Local-first AI inspection

AI traffic is inspected locally. Product analytics and error reporting can be disabled in Settings.

SQLite local storage

All inspection events stored locally as structured JSON. Your audit logs stay on your infrastructure — ready for SIEM ingestion or compliance review.

Scanner source on GitHub

github.com/citrusglaze/citrusglaze-scan

Where CitrusGlaze fits in the security stack

Prompt-level DLP Agentic AI security AI governance and enforcement Adjacent to SSE / Security Service Edge On-prem / data sovereignty capable

SSE platforms route traffic through their cloud for inspection. CitrusGlaze inspects locally — no cloud routing required.

What we don't have (yet)

We believe transparency builds trust. Here's where we are today.

No SOC 2 Type II certification

On our roadmap. Our scanner is open source (MIT) and available for your security team to audit directly.

No FedRAMP authorization

Not applicable to our current deployment model. AI inspection is local; Pro/Enterprise cloud sync is optional.

No 24/7 support team

Direct engineering support with response within 24 hours. You talk to the people who built the product.

No SLA guarantees

We're early-stage. Our uptime record is excellent but we don't offer contractual SLAs yet.

We're early-stage and transparent about it. Our code is the proof.

Responsible disclosure

Found a security issue? We take vulnerability reports seriously. Please contact us directly.

[email protected]

Ready to evaluate CitrusGlaze?

Start with the free scanner, or talk to us about your security requirements.