Product

Your application sends its prompt to Guarrix instead of directly to your LLM provider. Guarrix is designed to scan the inbound request, apply detection and your configured rules, then forward the cleaned prompt to your chosen LLM. The response comes back through Guarrix, is scanned again on the way out, and only then reaches your application. Both directions are inspected; nothing is stored in either direction.

Personally identifiable information (names, emails, phone numbers, IBANs, national IDs, and more), prompt injection and jailbreak attempts, and any custom entity or keyword rules you define — each rule can be scoped to inbound, outbound, or both directions, with its own redact, mask, or block action.

Guarrix is designed as a drop-in proxy: point your existing OpenAI-compatible API calls at your Guarrix endpoint instead of your LLM provider directly, using your Guarrix API key. Supported providers: OpenAI, Anthropic, Mistral, or any OpenAI-compatible endpoint you already use.

Free and Starter plans run on a shared, logically isolated instance with strict per-tenant API key separation. Pro and Business plans get a pool-isolated instance with no noisy-neighbour risk. Enterprise customers can choose a fully dedicated instance, a specific region, or a self-hosted deployment.