Our clients trust us with their most sensitive compliance data. With MATVIS, we guarantee that this data is protected at every step.
Everyone filters the prompt and the response.
MATVIS checks every tool call, every argument and
every result before it happens, and on whose authority.
Everyone else stops at the surface.
We are built
for depth.
Everyone filters the prompt and the response. MATVIS checks every tool call, every argument and every result before it happens, and on whose authority.
In the product
Runtime Security for
Every interaction, every direction, one place. From the first prompt to the last tool call, MATVIS covers all of it.
Book a demoTopics01 / 05
Guidelines are optional.
Enforcement isn’t.
Define what your AI may and may not talk about. Context-aware guardrails for every topic you set, enforced on input and output, so model behaviour follows your rules and not the user’s.
Vault02 / 05
The model never sees
the data.
Data loss prevention and masking for AI traffic. Personal and sensitive data is masked in the PII vault before it reaches the model and restored in the response. The model never sees it, the user never notices.
Identity03 / 05
Least privilege
for something that improvises.
An agent gets an identity: which tools, which arguments, whose authority. What it may not do, it cannot do, however it was prompted.
Tool results04 / 05
What comes back
is untrusted input.
Documents, web pages, API responses: the agent asked for them, so it trusts them. That is where hidden instructions and malicious content get in. MATVIS checks every result before the agent reads it, including everything an MCP server returns.
Evidence05 / 05
Focus on real threats.
Not noise.
Threats are stopped, not flagged afterwards. Every decision is kept as evidence, with the check that fired and the configuration it ran under, and streams into Splunk or whatever SIEM you run.
Where it runs
Your data stays in your environment.
MATVIS runs where you run: as SaaS hosted in the EU, in your cloud, or on-premise, air-gapped if you need it. No prompt or document leaves your environment. You keep control of data handling, auditability and compliance scope.
01
SaaS, EU-hosted
Hosted in the EU. Ready to use.
02
Your cloud
Your account, your region, your keys.
03
On-premise
Docker or Kubernetes in your data centre.
Two ways to start
Start small.
Commit with confidence.
01
Proof of risk
See your AI risk without changing anything. We analyse representative interactions and show you data exposure, unsafe prompts and policy gaps, with concrete recommendations.
Low effort. No production impact.
02
Proof of concept
Deploy in a controlled pilot with your real AI applications and workflows. Evaluate alerts, policies and processes in your own infrastructure, without the commitment.
Real-world validation. Technical evaluation only.
Ready when you are.
Tübingen, Germany · ISO 27001 certified
With MATVIS as our technology partner, our customers benefit from enterprise AI while enforcing GDPR compliance at the same time.
MATVIS runs entirely on our own infrastructure, responds orders of magnitude faster than anything else we measured, and protects more reliably. We did not expect to get all three at once.
Questions we get
Before
the call.
Straight answers to the questions that come up in every first conversation. If yours is not here, it is a good reason to book one.
Does MATVIS use an LLM to make decisions?
No. There is no language model in the decision path. Purpose-trained classifiers score every message, and the same input produces the same result every time. That is what makes a decision reproducible and auditable.
How is MATVIS deployed?
As containers, on industry-standard tooling. Either as SaaS hosted in the EU, or on-premise on Docker or Kubernetes in your own infrastructure, including fully air-gapped. Same firewall, same policy engine, same speed in every option.
Can it run air-gapped?
Yes. Air-gapped is on-premise with no outbound connectivity. The classifiers, the event database and the dashboard all run inside your environment, and the licence is verified locally.
How is this different from cloud provider guardrails?
Cloud guardrails screen the chat path, on their own models, inside their own cloud. They do not see which tool an agent calls, with which arguments, or what comes back. MATVIS checks both paths and runs where you run.
Which models does it work with?
Every model, every provider, one policy. OpenAI, Anthropic, Azure OpenAI, Google, and self-hosted runtimes such as vLLM or Ollama. Swapping the model behind MATVIS does not touch the client.
Where do my prompts go?
On-premise, nowhere: prompts, documents and tool payloads never leave your environment. In the EU-hosted SaaS they are processed in the EU under European law, for the check and nothing else. In both cases the vault masks personal data before a request reaches the model provider and restores it on the way back.
Is it audit-ready?
Yes. Every result carries the check that fired, the score, the identity and the configuration version it ran under. That is the audit trail the AI Act, NIS2 and GDPR ask for, without forensic guesswork after the fact.
Does it plug into my SOC?
Events stream to your SIEM, Splunk or whatever you run, through one configuration with live event flow. Threats are stopped at the moment they happen; your SOC sees the evidence and can overrule a result through the API.
Which regulations does this help with?
The EU AI Act, NIS2, GDPR and DORA all ask for controls around AI systems and for evidence that they were applied. MATVIS is the control, and every event is the evidence.