Security & EU hosting

European AI. With only the access you give them.

zwrm agents work in the tools your team uses every day. This page explains where they run, how they are kept apart, what they can reach and how you check what they did.

Hosted in Europe. By a German company.

Where your data is kept

  • Agent workspaces and files EU
  • Chats with your agents EU
  • Run log and audit log EU
  • Credentials for your tools EU, encrypted
  • Requests to the model Depends on the model ↓

Self-hosted, zwrm keeps all of this on your own servers.

Data centers

European infrastructure

Agents run on European infrastructure, in data centers owned and operated by European companies. What zwrm stores for you is stored and processed in the EU.

Company

A German company

zwrm is run by GTN Group AG in Freiburg. As an EU-registered company, it is not subject to the US CLOUD Act and does not depend on US cloud providers.

Each agent works in its own microVM. A fresh machine for every run.

Agents run code, open files and call tools. On zwrm, every run and every chat starts its own Firecracker microVM, a small virtual machine with its own kernel. When the work is done, the machine is destroyed.

  • Separated by hardware virtualization, not by containers that share one kernel.
  • A new machine for each run or chat, destroyed when it finishes.
  • The agent’s workspace is kept between runs, so its files and notes carry over.
  • Credentials for your tools never enter the machine. It gets a token for this session only, revoked when the session ends.

Only the access you give. Under the agent’s own name.

You decide which tools each agent may use and which account it signs in with. Connecting an account is always a step for a person, never for an agent.

Access

Scoped per agent

Attach only the tools an agent needs, down to single actions. Credentials stay encrypted in the zwrm gateway. Detach a tool or disconnect the account, and the agent loses access.

Identity

Its own account

In Microsoft 365, Notion and GitHub, each agent signs in with its own account, never with yours. What that account may see and edit is all the agent can reach there, and its replies carry its own name.

Every run is logged. Down to each tool call.

The dashboard shows what started each run, which tools the agent called with which account, and what came back. Export the audit log as CSV or JSON for your records.

Choose the model for each agent. And see where it runs.

Claude and Codex agents use your own Anthropic or OpenAI account. Agents on the zwrm runtime use open-weight models such as Mistral, GLM and Kimi, provided by zwrm. Every model is labeled with where it runs.

Where your messages go

To answer, an agent sends your message to the provider of the model it uses. Anthropic and OpenAI models run outside the EU and are labeled that way.

EU models only

An organization setting hides models that run outside the EU from the zwrm model catalog and refuses requests to them. Admins switch it on in the dashboard settings.

Your data stays yours

You keep all rights to your data. zwrm accesses your application data only where needed to provide the service, answer a support request or meet a legal obligation. When your contract ends, it is deleted after 30 days.

For your security review. Self-hosting, DPA and our policies.

Tell us what lands on your desk every week. We’ll show you an agent doing it.

A demo with our team, in English or German. Bring one recurring task and the tools it touches.