AI Data Security Gateway
Every prompt, screened before it leaves your network. Every response, screened before it touches your infrastructure.
Introducing Phocion. A model-agnostic proxy that sits between your applications and any LLM provider — masking sensitive data in real time and screening for prompt injection, without changing how your team calls the model.
The problem
Teams are wiring LLMs directly into products and internal tools faster than their data-governance policies can keep up. Customer PII, PHI, and proprietary business data routinely end up in prompts sent to third-party model providers — with no consistent layer checking what goes out, or what comes back.
How it works
Drop-in proxy
Model-agnostic — sits in front of whichever LLM provider you already use, no application rewrite required.
Real-time masking
PII and PHI are detected and masked in-flight, before a request reaches the model.
Injection screening
Every request is semantically screened for prompt-injection attempts before it reaches the model.
Self-hosted / VPC
Runs inside your own infrastructure — sensitive traffic never has to leave your network boundary.
What's real, right now
- Runs as an inline proxy alongside your existing infrastructure, not a round-trip to a third-party scanning service.
- Designed to reduce token overhead via semantic caching — savings are directional, not yet a measured, published figure.
- Built with SOC2/HIPAA-oriented controls in mind. Phocion is not currently SOC2 certified.
Status
Phocion is in active development, running technical pilots. This page exists so the claims made about it elsewhere have something real to point to.