Skip to content

Your data, someone else's infrastructure

Vidai runs in your infrastructure. Your prompts, your provider keys and your records stay there. Use cloud providers, your own models on your own machines, or both. The licence is checked offline, and nothing phones home.

The problem

Your security team has one question about every new AI tool: where does our data go? For a tool that sits in the path of every prompt, the answer has to be "nowhere new", or the review never ends. And for some teams, some data must never leave a region, or the building.

Why it stays unsolved

Many AI tools are hosted services: to use them, every prompt and every provider key passes through someone else's infrastructure, in someone else's region, under someone else's retention. Each one is a new data processor to assess, a new agreement, a new place for a leak. The tool meant to control the data becomes another place the data goes.

How Vidai handles it

  • Self-hosted, all of it. The gateway, the Control room and their database run where you run them. VidaiGuard Enterprise runs beside the gateway. What Enterprise includes depends on the features your licence includes.
  • Your providers, your keys. Calls go to the providers you connect, with your own provider keys, held centrally and never shown in full.
  • Your own models. Connect models on your own machines, such as Ollama or any OpenAI-compatible server, beside the cloud providers, and route sensitive work to them.
  • Where each provider runs is on record. Region rules Enterprise keep calls in the areas you choose, and only models on your approved list are reachable at all. See Regions and data residency.
  • No phone-home. The licence is checked where Vidai runs, offline.

Models & routing › Providers: each provider, where it runs, and how many of its models are switched on

What changes for you

  • The security review is short. Data flows from your apps to your gateway to the providers you chose, and nowhere else.
  • Sovereignty is a setting, not a project. Keep a team's calls in the EU, India or the US, or on your own hardware.
  • No concentration on one provider. Providers are swappable behind stable model names.
  • Proof: the architecture on The parts.