What never reaches us.

We set up and run your gateway, but we're never in the middle of it.

Never sent to us

  • What your staff and agents ask the AI, and what it answers, including any documents or tool results attached.
  • The keys to your accounts with OpenAI, Anthropic, Google and the other AI companies.
  • What a safety rule caught. We learn that a rule fired, never the text.
  • Your staff list. Names and emails stay in your cloud, and we only see an ID for each person. Your team still sees names.
  • Our outage is not your outage. Nothing of ours sits between your staff and the AI. If we go offline, your gateway keeps working on its last settings: budgets, routing and backups included.
  • Support only when you allow it. The access you approve is for running the gateway (installing, upgrading and monitoring it), never for reading requests. When you want help, you grant time-limited, read-only access.
  • Stored where you are. Usage records and your console's data are kept in your region.

Your security team will want the detail: our data map, security questionnaire answers and an architecture overview.Request our security pack

GDPR, the EU AI Act and your contracts.

What running the gateway in your own cloud means for the rules you work under.

GDPR

Prompts, answers and documents can contain personal data. With Intramura they go from your staff straight to the AI companies you have contracted, so we are not a processor of that data and don't join your chain of processors for it.

You decide where requests go, so personal data can go only to AI companies and regions you have approved, such as endpoints in the EU.

We do receive usage records, with each person as an ID rather than a name. For those we are your processor, under a data processing agreement.

EU AI Act

If you deploy a high-risk AI system, you must keep the logs it generates. With Intramura, every AI request leaves a usage record, which you can export to your own monitoring tools and keep.

The audit log shows who changed which rule, budget or permission, and when.

Data processing agreement

We sign a data processing agreement for the usage records we receive, with the list of companies that help us run our service.

Ask for the agreement

Security questions

Do my prompts ever reach Intramura?

No. The gateway runs in your cloud account, and requests travel from your applications to the model providers you enable. The access you approve is for running the gateway, never for reading requests or answers.

What does Intramura receive?

Only what it takes to show you spend and keep the gateway running: for each AI request, who it belongs to (as an ID, never a name), which AI model answered, what it cost and whether it worked. Never the request itself, the answer, or your AI accounts' keys. The full list is in our security pack.

What staff data do you hold, and why?

Nothing from your staff list that identifies a person. People who sign in to the console do so through your identity provider, so we see their name and work email. Your staff list (names, work emails, teams and managers) stays in your cloud account, and we only ever see an ID for each person. Your own team still sees names. The list is used to put an owner on spend, apply team budgets and switch off leavers' access. We never ask for salaries, contracts, performance reviews or personal contact details, and we can't change anything in your systems.

Is the gateway open source?

It is open source under the Apache 2.0 licence. We share the full stack, including the project, in our security pack and in the demo.

Find out where your AI budget goes.

Tell us how your teams use AI today and which cloud you run on. We will show you what spend control looks like in your own account.