Secure LLM proxy · scan → forward → scan

LLM Gateway

Put a security firewall in front of any model. The gateway scans your prompt, forwards it to the provider/model you choose, then scans the response — masking leaked secrets/PII and (in block mode) stopping prompt-injection before it reaches the model or the response reaches your user.

Prompt to send through the gateway

System prompt (optional)

Policy

Everything runs; risks are reported.

Model & key (BYOK — never stored)

Without a key you get a clearly-labeled mock response (the scans still run). Groq is text-only for chat here.

Send a prompt to see the input scan, the model response, and the output scan.

The gateway is also available as an API: POST /api/gateway. Deterministic scans are authoritative; the model is the component being guarded. See the docs for the request schema.