GuarrixThe AI guardrail that blocks PII and prompt injection before they reach your LLM
Bidirectional guardrail between your application and any LLM. EU-sovereign. We can't leak what we don't store.
The problem
Every prompt you send to an LLM provider is a potential data-protection incident waiting to happen.
Personal data leaves your infrastructure
Every prompt your users type — including any name, email, or ID number they paste in — travels to your LLM provider's servers. Most teams have no filter in place before that happens.
Prompt injection is hard to catch by hand
Attackers hide instructions inside normal-looking input, trying to override your system prompt or extract data your model shouldn't reveal. Catching every variant yourself is a losing game.
No visibility into what actually left
Without a guardrail, you find out about a leak from a customer complaint or a data-protection inquiry — not from your own logs, because there's nothing recording what was sent.
How it works
Your app sends prompts to Guarrix instead of directly to the LLM. The guardrail validates both directions before anything moves.
EU sovereignty
Dutch jurisdiction, no US parent, no CLOUD Act exposure.
Bidirectional
Every prompt and every response is validated before it moves.
RAM only
Content is processed in memory and discarded. Never logged, never cached.
What's included
Everything you need to run a guardrail in production, not just a demo.
Custom rules per tenant
Define your own regex or keyword patterns, choose redact, mask, or block, and scope each rule to inbound, outbound, or both.
Metadata-only audit log
Every request is logged — token counts, detected entity types, action taken, latency — without ever storing the prompt or response content itself.
Works with any LLM provider
OpenAI, Anthropic, Mistral, or any OpenAI-compatible endpoint. Switch providers without changing your detection rules.
One dashboard for every tenant
Manage entities, custom rules, API keys, and billing for every tenant from a single place.
Try the guardrail
Paste text. We detect PII and return an anonymised result. No signup, no storage.
Try the guardrail
Paste text. We detect PII and return an anonymised result. No signup, no storage.
Guarrix vs. the alternatives
Where Guarrix fits compared to other ways to protect LLM traffic.
| Alternative | Where Guarrix differs |
|---|---|
| AWS Bedrock Guardrails | AWS lock-in, US jurisdiction, enterprise pricing, inbound checks only — Guarrix is EU-jurisdiction, bidirectional, and provider-agnostic. |
| Protecto / Nightfall | US companies with CLOUD Act exposure — Guarrix is a Dutch company operating entirely under EU law. |
| LiteLLM (open source) | Requires self-hosting and setup, no managed service, no dashboard, no outbound scanning — Guarrix is a managed service covering both directions out of the box. |
| Anonify.nl | Document redaction only, not a full AI gateway — Guarrix sits in front of your LLM calls end-to-end. |
Ready to protect your LLM traffic?
Start free — no credit card required.
Get started