AI privacy & guardrails

Use AI without sending the wrong data.

AI requests can contain more data than intended.

Guarrix checks each request before it reaches your AI provider, applies your configured rules, and checks the response on the way back. You keep using your existing AI setup while adding control over what is sent.

Get started Try the sandbox

Start free. No credit card required.

Test rules before you activate them.

Changes to data-handling rules should be tested before they reach production.

The Guarrix sandbox lets you run a sample through your configured guardrail and inspect the result before activating the policy.

what Guarrix detected
what was redacted
which filters were triggered
what was sent to the model
what the model returned

Try it yourself

Paste some text below and see what Guarrix catches — no signup needed. When you're ready, start small: the free plan includes 500 requests a month, no credit card required.

Filter only. This shows the filter only — no AI model is involved. Your message is never sent anywhere except our own PII detector.

Try a live filtered chat

See exactly what leaves as PII-safe, message by message. No signup, no storage.

Filter only. This shows the filter only — no AI model is involved. Your message is never sent anywhere except our own PII detector.

Try an example:

AI requests can contain more data than expected.

A user pastes customer information into a prompt.

A developer sends production logs to a model.

An internal application includes an employee name, phone number, account number or internal identifier in an API request.

That data is often included as part of the request without anyone explicitly deciding to send it.

Once the request reaches the AI provider, that control point has already passed.

Your AI provider is only part of the picture.

You can choose an EU provider. You can sign a DPA. You can configure retention. You can restrict which models employees are allowed to use.

Those controls are useful. But there is another question to answer:

What exactly are your applications sending to the model?

Guarrix adds a control point at that boundary.

ApplicationGuarrixAI provider

Privacy entities screen listing PERSON, EMAIL_ADDRESS, PHONE_NUMBER, IBAN_CODE and other entity types with action, state and enabled controls

Decide what your application is allowed to send.

Guarrix detects common sensitive data types before they reach the model.

Names. Email addresses. Phone numbers. IBANs. Credit-card information. IP addresses. Locations. And more.

For each type, you decide what happens.

Allow it. Replace it. Hash it. Redact it. Or block the request.

Your application stays in place.

Your AI provider stays in place.

Guarrix applies the rules between them.

Your sensitive data isn't always standard PII.

CLIENT-43821PROJECT-FALCONINTERNAL_SECRET

Those values may not be recognised by a standard PII detector.

They can still be sensitive within your organisation.

Guarrix lets you create custom regex and keyword rules for business-specific patterns that also need protection.

Custom rules screen with regex-based rules for password assignment, API tokens, bearer tokens and database credentials

Security filters screen with preset selector and grouped filters for identity and regulated data, secrets and credentials, business-sensitive data and AI security

Add broader protection without building every rule yourself.

Guarrix is not limited to personal data.

It can also help detect secrets, credentials, business-sensitive patterns and AI-specific risks before requests reach a provider.

Use built-in filter groups for common categories such as identity and regulated data, secrets and credentials, business-sensitive data, and AI security.

Detect only. Block. Or combine built-in filters with your own custom rules.

Add a control layer without replacing your AI stack.

Keep Mistral. Keep OpenAI. Keep Anthropic. Keep your self-hosted model.

Guarrix is not another AI provider. It sits between your application and the provider you already use, applying your policies before requests are forwarded and checking responses on the way back.

Your application

Guarrix

DetectPolicies

AI provider

Response

Guarrix

Verify response

Your application

See what Guarrix is doing in production.

Guarrix gives your team visibility into requests, detections, blocks and latency so you can see how the guardrail behaves in production. The goal is to make the control layer observable without turning it into another conversation archive.

Guarrix dashboard showing recent requests, request breakdown and top detection types

AI governance gets harder as usage grows.

At first, one developer uses an AI API.

Then another application starts using it.

Customer support adds a model.

Another team connects a different provider.

An internal agent starts calling tools.

A new MCP integration appears.

As AI usage grows, the number of places where data can leave your environment grows with it.

Guarrix gives those systems a common place to apply data-handling rules before requests are sent.

App 1Support botAgentMCP tool

Guarrix

Know where your AI data goes.

Guarrix adds a privacy and policy layer to your existing AI setup.

EU hosted

Your Guarrix environment can run within European infrastructure.

No conversation storage

Guarrix is designed to inspect and protect requests without becoming another conversation archive.

Provider independent

Use Guarrix with the AI provider or model that fits your workload.

Policy controlled

Your organisation decides what gets allowed, masked, redacted or blocked.

Add controls where your applications connect to AI.

Internal AI applications

Protect employee and business data before internal tools call an AI model.

AI-enabled SaaS products

Add a privacy layer between your users, your application and your AI providers.

Agents and automated workflows

Apply consistent data policies before autonomous systems communicate with external models or tools.

Regulated environments

Add a technical control point for teams that need visibility and policy enforcement around AI traffic.

There are several ways to protect AI traffic.

You can build filters yourself, depend on your AI provider's controls, run a standalone PII tool, or put a dedicated guardrail layer between your applications and your models.

Guarrix is built specifically for that last role.

AWS Bedrock Guardrails

Tied to AWS and its model catalogue. The company behind it is US-headquartered regardless of which AWS region you pick — Guarrix is a Dutch, EU-jurisdiction company you can put in front of any provider, including AWS.

Protecto / Nightfall

US companies, so still subject to US jurisdiction (such as the CLOUD Act) even when you use their EU infrastructure — Guarrix is a Dutch company operating entirely under EU law.

LiteLLM (open source)

A self-hosted routing gateway with its own dashboard, but PII and prompt-injection detection comes from third-party guardrail plugins you wire in yourself — Guarrix ships that detection built in, as a managed service covering both directions.

Anonify.nl

Document redaction for FOI and public-records workflows, not a live AI request/response gateway — Guarrix sits in front of your LLM calls in real time, not just documents.

Lakera Guard

An attack-focused AI security platform, now part of Check Point, aimed at enterprise security teams — Guarrix is a simpler, EU-hosted guardrail layer for teams that don't need a full enterprise security suite.

Protect AI

Was an independent AI/ML security company; now part of Palo Alto Networks' Prisma AIRS platform, covering the broader AI development lifecycle — Guarrix is scoped specifically to the request/response layer in front of your LLM calls.

Build it yourself

You can build your own checks into your app. This gives you full control, but you also have to build, test and maintain everything yourself. As new risks appear, your protection needs to change with them. This can work well if you have the time and people to manage it.

Guarrix does this work for you, so your team can focus on your app.

Use the protection from your AI provider

Some AI providers include their own safety tools. They can be useful, but your protection can become connected to that provider. If you change AI providers later, you may also have to change how you protect your app.

Guarrix keeps the protection separate. You can choose the AI provider that works best for you and change providers later without starting over.

Only remove private data

You can add rules that remove names, email addresses and other private information before sending something to AI. That solves one part of the problem. But private data is not the only risk. Someone may try to trick your AI into ignoring its rules. The AI itself can also return something you do not want your users to see.

Guarrix checks both what goes in and what comes back.

How does Guarrix compare?

Guarrix is not the only product that helps companies use AI more safely. Different products are built for different needs. Here is a simple overview to help you compare them.

Guarrix

Guarrix is made for businesses that want a simple protection layer between their app and their AI provider. Your messages are checked in Europe and the full content is not stored. You can keep using your own AI provider and change providers without replacing Guarrix.

Lakera Guard

Lakera Guard is an established AI security product with a strong focus on protecting AI applications from attacks. It can be a good fit for organizations looking for a larger AI security platform and enterprise security features.

Protect AI

Protect AI offers a broader set of tools for securing AI systems. It can be a good choice for organizations that want security across more parts of their AI development and infrastructure.

Why choose Guarrix?

We don't believe one product is the right choice for everyone. Guarrix is built around a few simple ideas:

Keep your AI provider — switch providers whenever you want, without replacing your protection.
Keep your messages private — the full content is never stored, only metadata.
Keep your data in Europe — the checking itself runs on our servers under EU law.
Protect both directions — what you send and what the AI sends back.
Start small — try the sandbox free, no signup, then add a plan when you're ready.

The goal is simple: give your AI an extra safety check without making your application more complicated.

Start with one workload and expand from there.

You do not need to redesign your AI stack to start using Guarrix. Connect an application, configure the rules and test the flow before adding more workloads.

View pricing

Common questions

A few quick answers — see the full FAQ for more.

FAQ

Use AI while keeping control of the data.

Add a privacy and policy layer between your applications and the AI providers you already use.

Start free. No credit card required.