How it works

Local-first compliance. Zero data exposure.

HoundShield sits between your team and every AI provider. Everything is scanned on your hardware — not ours. That's the only way to guarantee your CUI never becomes a DFARS 7012 spill.

01

Change one URL

Point any OpenAI-compatible client (ChatGPT, Copilot, Claude, Cursor) at your HoundShield proxy URL. No agents. No installs. Live in under five minutes.

02

Every prompt scanned locally

HoundShield intercepts each request on your own hardware. Sixteen CUI/PII/PHI engines run in under 10ms before the prompt reaches any AI provider. Your data never touches our servers.

03

Block, quarantine or pass

Clean prompts pass through untouched. Risky prompts are blocked or held for human-in-the-loop review — with the exact match (CAGE code, MRN, secret) highlighted.

04

Sign the evidence

Every decision is written to a SHA-256 tamper-evident audit log. Export an assessor-reviewable PDF mapped to all 110 NIST 800-171 controls whenever your assessor asks.

Scan free

FAQ

Deployment questions, answered

Still have questions? Talk to a compliance engineer — we respond within 4 business hours.