HoundShield

The local-only AI compliance firewall.

Scan every AI prompt for CUI, PII, PHI, and secrets in <10 ms — on your own network. Nothing leaves the building.

Website · Pricing · Compare · Contributing · Report a Vulnerability

---

Why HoundShield

Every cloud AI data-loss tool — Nightfall, Strac, Microsoft Purview — scans your prompts by first sending them to the vendor's servers. For a defense contractor, that transmission is itself a DFARS 252.204-7012 CUI spill. You cannot scan Controlled Unclassified Information for compliance by violating compliance.

HoundShield runs the scan locally. Prompt content never leaves the customer network. That is the entire pitch — and the only architecture that is honest about CUI.

Built for Jordan — the IT Security Manager at a 50–250 person DoD contractor staring down the CMMC Level 2 deadline of November 10, 2026. ~80,000 contractors need it. A few hundred are certified today. The clock is the product.

---

What it does

Capability	Detail
🛡️ 16 detection engines	CUI, PII, PHI, and secret patterns scanned inline on every prompt
⚡ <10 ms local scan	Pattern matching runs on-prem — no round trip to a vendor cloud
📊 SPRS scoring	Live score across all 110 NIST 800-171 Rev 2 controls
🔗 Tamper-evident audit	SHA-256 hash-chained log — every decision provable, nothing editable after the fact
📄 C3PAO-ready PDF	One-click assessment export your assessor can actually use
🚫 Inline policy	Block, redact, or warn — enforced before the prompt reaches the model

---

How it works

flowchart LR
    A[Employee] -->|prompt| B(AI tool · ChatGPT · Claude · Copilot)
    B -->|HTTPS| C{{HoundShield Proxy<br/>local · on-prem}}
    C -->|16-engine scan · &lt;10ms| D[/Classifier/]
    D -->|clean| E[(LLM provider)]
    D -->|CUI / PII / PHI / secret| F[Block · redact · audit · alert]
    C -.->|prompt content never leaves the network| C

Loading

The proxy sits between your people and whatever model they use. It inspects the request locally, applies policy, writes a hash-chained audit entry, and only then — if the content is clean — forwards it. Detection happens where your data already lives.

---

Deployment modes

Mode	Hosting	CUI-safe?	Use it for
A	Vercel-hosted (houndshield.com)	❌ No — demo only	Evaluation and sales demos. Never send real CUI through Mode A.
B	Self-hosted Docker	✅ Yes	Production defense workloads on your own network
C	Air-gapped	✅ Yes (IL-5+)	Classified or fully disconnected environments

Mode A is a hosted demo and is not CUI-safe. It exists so buyers can try the product. Production CUI handling requires Mode B (self-hosted) or Mode C (air-gapped).

---

Quick start

Mode B — the proxy (the actual product):

cd proxy
cp .env.example .env        # set your license + upstream config
docker compose up -d        # HTTPS intercept proxy comes up locally

The dashboard / control center:

cd compliance-firewall-agent
npm install
npm run dev                 # http://localhost:3000

Verify before shipping:

cd compliance-firewall-agent
npm run build               # must pass
npm test                    # 400+ tests, must stay green

---

Tech stack

Application — Next.js 15 · React 19 · TypeScript (strict) · Tailwind CSS · Framer Motion · Recharts · Supabase (auth + Postgres) · Stripe.

Proxy — Node.js HTTPS intercept · deterministic pattern scanner (16 engines) · OODA-loop rate/anomaly tracker · Docker / docker-compose · air-gap-capable.

---

Compliance coverage

• NIST SP 800-171 Rev 2 — all 110 controls, mapped and scored
• CMMC Level 2 — readiness and SPRS self-assessment support
• DFARS 252.204-7012 — local-only architecture avoids the scan-time CUI spill
• HIPAA — PHI detection patterns
• SOC 2 — control mapping

HoundShield is compliance tooling, not a certification. C3PAOs assess; HoundShield prepares the evidence, the score, and the audit trail.

---

Project structure

compliance-firewall-agent/   Next.js app — dashboard, API, Brain AI, classifier
proxy/                       HTTPS intercept proxy (the product)
  patterns/                  16 CUI/PII/PHI/secret detection engines (extend, never reduce)
  ooda/                      observe → orient → decide → act rate engine
supabase/                    database migrations
docs/                        PRD, roadmap, setup, compliance references

CI enforces a hard floor: the Compliance Pattern Guard fails any change that drops the engine below 16 patterns. The compliance engine is never silently degraded.

---

Pricing

Free → Pro → Growth → Enterprise → Agency. Current tiers and pricing live at houndshield.com/pricing.

---

Contributing & security

• Contributing — see CONTRIBUTING.md. Builds must pass and the pattern count must never drop below 16.
• Security — found a vulnerability? Do not open a public issue. Follow the disclosure process in SECURITY.md.
• Code of Conduct — CODE_OF_CONDUCT.md.

---

License

MIT © HoundShield.com

Scan locally. Spill nothing.