Your clients' AI code.
Sovereign, isolated, under your brand.

Every Guard runs in its own MicroVM (Kata Containers) with its own guest kernel via hardware virtualization (KVM). That is the strongest practical isolation level for untrusted code, the same approach used by global hyperscalers. Every Guard is separated from other projects and from the host kernel, and every access is logged. Your data stays exclusively in Germany.

No, and that's by design. endless guard doesn't evaluate your code's business logic or functionality. What we secure is the infrastructure, data flows, and the runtime environment. Because every deployment runs hardware-isolated, even unexpected code behavior stays contained within its own Guard and can't spread to the host or other projects.

Yes. Every project gets its own isolated Guard with a dedicated MicroVM. The Basic plan includes 5 Guard units, which you distribute freely across Guards — five small Guards with one unit each, for example, or two medium ones plus a small one. Additional units can be added anytime at €29.80 €14.90 / month.

The Guard unit is our resource and billing size. One Guard unit comprises 2 vCPU, 2 GB RAM, and 20 GB of local NVMe storage. A Guard is always exactly one MicroVM, and depending on its size it consumes a fixed number of units from your quota: S with 2 vCPU / 2 GB RAM / 20 GB (1 unit), M with 4 vCPU / 4 GB RAM / 40 GB (2 units), L with 8 vCPU / 8 GB RAM / 80 GB (4 units). Only the sum of units is capped, not the number of Guards: with 5 units you can run five S Guards, two M Guards plus one S Guard, or one L Guard plus one S Guard. Guards larger than 4 units are provisioned on dedicated hardware under the Enterprise plan. vCPU describes a burst share on shared physical cores and is capped per Guard so that individual workloads — build processes, for example — cannot degrade the performance of other Guards. RAM and local storage, by contrast, are firmly allocated.

You decide that deliberately, through where the data lives. A unit's local NVMe storage is meant for code, runtime, dependencies, and caches, and it is not replicated: after a restart or a node failure, we automatically restore your Guard from its last deployment image. Anything that has to survive a redeployment — databases, uploads, generated files — belongs on a persistent volume. That is replicated storage, available as an add-on at €0.25 net per GB per month in 10 GB increments. This way you pay for replicated storage only for the data that actually needs it, not for every dependency directory.

endless guard automatically logs tamper-proof access logs and complete audit trails at the platform level. This evidence forms the compliance layer you need for your own documentation and your clients': for GDPR and NIS2, and as the technical foundation for further requirements like the EU AI Act. Responsibility for your application's substantive compliance stays with you; we provide the provable infrastructure underneath it.

endless guard is primarily optimized for local workflows like Claude Code. That said, you can deploy any containerized AI application that communicates over standard interfaces like MCP or REST.

Your local agent connects directly to our API via the Model Context Protocol (MCP). A single command is enough to transfer your code, cryptographically signed, into a new, isolated MicroVM.

All Guards run in ISO 27001-certified data centers in Germany. One Guard unit comprises 2 vCPU, 2 GB RAM, and 20 GB of local NVMe storage. Depending on its size, a Guard uses 1 to 4 units. Hosting and maintenance are fully included in the price, so you don't need your own server. We guarantee strict GDPR compliance and digital sovereignty.

If the goal isn't reached by December 31, 2026, the project will be discontinued and all payments will be automatically and fully refunded via Mollie on January 1, 2027.