Security & data isolation

Where your data lives, and who can't touch it.

Serious buyers ask the same five questions. Here are the answers in plain language, including the honest parts a sales page usually hides.

The architecture, in one breath

One company · one instance

Every customer gets their own sealed instance with its own database. Your leads, customers and documents never share storage with another company. There is no shared pool to leak across, because there is no shared pool.

Your keys · your model accounts

The AI runs on your own model-provider accounts, billed at cost. Your business data goes to the model under your agreement with that provider, not through a middleman account mixed with other companies' traffic.

The approval queue

Nothing external ships without a human stamp. Every outreach email, social post and deal move waits in a queue until a person at your company approves it. AI drafts; a human decides. A design constraint, not a setting.

EU jurisdiction

Built and operated by Byggarn AB in Lund, Sweden, under GDPR and Swedish law. For your instance's data we act as data processor under a written agreement; you stay the controller. AI-written messages are disclosed as AI per the EU AI Act.

No training on your data

We do not train models on your business data. Your instance's memory exists to serve your company and no one else.

The honest part

We are a young company and we act like it. We do not yet hold a SOC 2 or ISO 27001 certificate, and we won't pretend otherwise with a badge wall. What we offer instead, today: the isolation architecture above, a signed data-processing agreement, and direct access to the people who built the system, not a support tier.

Every instance is set up by hand, which means onboarding is deliberate and slow by SaaS standards. We consider that a security feature.

· KLADDS, Lund

Questions we expect on the call

Can you see our data?

As operator we can access an instance for support and maintenance, under the DPA, logged. We don't browse customer instances, and your data never appears in another customer's instance.

What happens if we leave?

Your instance's data is exportable and the silo is deleted on termination, per the DPA's retention terms.

What if the AI writes something wrong?

Then it sits in the approval queue where your editor catches it. That is the whole point of the queue: the cost of a bad draft is zero if it never ships.

Subprocessors?

Listed in the DPA you sign, and the model providers are the ones you choose, on your keys.