The Defensible AI Platform
A reverse proxy for enterprise AI traffic. Every prompt scanned for PII. Every interaction signed with cryptographic proof. Deploy in 30 minutes. No employee behavior change required.
Most won't admit it. You have no visibility into what data they're sending to external AI models. You can't answer a regulator's question about your AI usage.
Social security numbers, patient data, financial records, source code. Once it reaches an external model, you've lost control. In 2023, Samsung engineers pasted proprietary semiconductor code into ChatGPT — the company only discovered it after an employee report.
When a regulator asks "How is your organization governing AI usage?" — what do you hand them? Email threads? Meeting notes? Nothing?
Banning AI pushes usage to personal devices outside your perimeter. Approving specific tools without monitoring gives you compliance theater, not compliance.
Route AI traffic through a single control point. Employees use their tools normally. You see everything.
Core high-risk PII classes scanned on every prompt — SSN, credit card, email, phone, IP, DOB, medical records, bank accounts, passport, driver's license — with an extensible rule engine for custom patterns.
Block, redact, warn, or allow — per department, per user, per model. Custom keyword blocking. Max prompt length limits.
Every interaction signed with SHA-256 + HMAC. Merkle tree integrity. Tamper-evident. The same evidence infrastructure used by our Decision Council.
Track AI spend by department, team, project, and user. Per-model pricing for OpenAI, Anthropic, Google. Know exactly what AI costs you.
Self-hosted. Your infrastructure. Your keys. Air-gap capable. No cloud dependency. Data never leaves your perimeter.
For API-based model usage: change one environment variable. Point OPENAI_API_BASE at the gateway. Done. Browser-based AI tools require additional configuration (extension or network proxy).
The honest comparison.
| Capability | Microsoft Purview | Zscaler / Netskope | Helicone / LiteLLM | CendiaGateway |
|---|---|---|---|---|
| AI traffic monitoring | Yes (Azure only) | Partial | Yes | Yes (any provider) |
| PII detection in prompts | Yes (deep DLP) | Limited | No | Yes (extensible) |
| Policy enforcement | Yes | Yes | No | Yes |
| Cryptographic evidence packets | Not available | Not available | Not available | SHA-256 + HMAC + Merkle |
| Signed compliance artifact | Not available | Not available | Not available | AI Manifestâ„¢ |
| Self-hosted / sovereign | No (Azure) | No (SaaS) | Yes | Yes (air-gap capable) |
| Open-source | No | No | Yes | Yes (Apache 2.0) |
| AI decision governance | Not focus | Not focus | Not focus | Yes (The Council) |
For API-based AI usage (developer tools, internal apps, Copilot integrations) — one environment variable change.
The compliance artifact a CISO hands to an auditor.
30-minute technical briefing. We'll show you the gateway, the PII scanner, and the AI Manifest — using your own sample data.