Platform

Beyond the desktop.
One architecture. Four use cases.

The kernel, security model and network stack we built for a sovereign desktop turn out to be exactly the architecture the market needs for three harder problems.

Where each use case stands. Desktop · alpha is a working operating system you can boot today (QEMU and hardware). The Agent Runtime, OT/Embedded profile and Sovereign Compute node roadmap are built on the same kernel: their foundations — capability isolation, signed audit logging, WASM sandboxing with memory caps, local-model integration — already run and are tested in the kernel today, while the packaged products are on our roadmap (Horizon E). The terminal panels below are illustrative of intended behaviour, not captured system output.

EuroAgent Runtime roadmap · foundations built

AI agents that do real work — capability-isolated at kernel level, with an immutable audit trail, running local models. No mandatory cloud. No data leaving your organisation.

  • Each agent runs in its own WASM sandbox with declared capabilities
  • EuroGuard enforces network, filesystem and syscall policy in real-time
  • Every action logged in an Ed25519-signed, immutable audit chain
  • Local model by default (Ollama-compatible) — works completely offline
  • GDPR-compliant by design: data never leaves the hardware boundary
  • EuroIPC for secure agent-to-agent communication
# EuroAgent loading document-review agent

[agent] verifying signature... ok
[cap] fs:read → /data/contracts/**
[cap] net:out → 127.0.0.1:11434 (ollama)
[cap] net:out → 0.0.0.0/0 DENIED
[cap] mem:max → 512 MB

[ok] sandbox provisioned — agent running

# audit trail (immutable, Ed25519-signed)
t+0.12s READ /data/contracts/acme-2026.pdf
t+0.34s INFER local:mistral-7b → ok
t+1.02s WRITE /data/output/summary.md
t+1.03s EXIT code=0 sandbox=destroyed

# data never left the machine.
Illustrative — intended behaviour of the EuroAgent Runtime, not captured output.

OT & Embedded Profile Roadmap profile

A stripped EuroOS profile for operational technology — SCADA, PLCs, industrial controllers, edge nodes. Secure by design. NIS2-compliant. Runs on hardware from 2010.

  • 75–95 MB RAM — no hardware replacement needed
  • EuroGuard: every process has exactly the rights it needs, nothing more
  • Own network stack — no unexpected outbound connections possible
  • Verify-before-execute: only signed firmware and software runs
  • Full audit log — every action traceable, NIS2 Art. 21 aligned
  • Air-gappable — zero mandatory external connectivity
# EuroOS OT — industrial gateway node

[boot] verified boot chain — ok
[fs] EuroFS mounted — encrypted
[net] firewall active — default deny

# allowed: SCADA protocol on local segment
[net] ALLOW 192.168.10.0/24:502 (modbus)
[net] BLOCK 0.0.0.0/0 — policy

# EuroGuard — process capabilities
[guard] scada-daemon: net:local fs:/var/scada
[guard] scada-daemon: net:out DENIED

[sys] uptime 847d mem 81/512MB procs 19
[nis2] Art.21 logging active — all events signed
Illustrative — intended behaviour of the OT profile, not captured output.

Sovereign Compute Node roadmap · offering

Hardware plus software — a fully managed compute node that runs EuroOS, configured and delivered by GoTrust. Organisations get full control over their data without managing the complexity.

  • On-premise deployment — data never leaves the building
  • Local AI inference (EuroAgent Runtime included)
  • Full-disk encryption, TPM-anchored keys, verified boot
  • Zero telemetry — nothing phones home, by design and by audit
  • GoTrust manages updates and security patches
  • Suitable for air-gapped and classified environments
# sovereign compute — attestation report

node euro-node-0042
kernel EuroOS v0.2.1 (reproducible build)
tpm PCR0-7 verified — boot chain intact
disk encrypted key:hardware-bound
telemetry none — blocked at kernel level

# running services
[ok] euro-agent-runtime (3 agents active)
[ok] ollama:mistral-7b (local inference)
[ok] euroguard (monitoring all)

[attest] report signed — EuroCA verified
# provable: this machine runs what it claims.
Illustrative — intended behaviour of a managed node, not captured output.

EuroOS Desktop alpha · working today

A full sovereign desktop for organisations that cannot trust foreign OS vendors. From boot to browser to office suite — every line of code is in Europe, readable and auditable.

  • Own bootloader, kernel, filesystem, network stack — no inherited code
  • EuroSuite: Writer, Calc, Impress — opens OOXML and ODF
  • EuroWeb browser with own HTML/CSS engine and EuroTLS 1.3
  • 75–95 MB RAM with full desktop running
  • 24 EU languages built in, multilingual screen reader
  • EUPL-1.2 — fully auditable by any government or institution
# EuroOS desktop — system status

[boot] UEFI → EuroKernel in ~1.6s
[fs] EuroFS — CoW, encrypted, checksummed
[tls] EuroTLS 1.3 — EuroCA verified
[apps] all binaries Ed25519-verified before exec

# running (all sandboxed)
EuroWeb pid=42 Protected
EuroWrite pid=43 Protected
EuroCalc pid=44 Protected

memory 91 / 16384 MB
procs 27
[guard] 0 policy violations today
Illustrative status panel — the desktop itself is real and bootable from /try/.
EuroAgent Runtime

The EU answer to
agent-first computing.

AI agents are becoming real infrastructure. The question is who controls the trust boundary — a US cloud provider, or your own kernel. Roadmap · foundations built & tested

Agent loaded
WASM module verified by Ed25519 signature
Capabilities assigned
Policy declares exact fs, net and memory rights
Agent runs
EuroGuard enforces policy on every syscall
Audit trail
Every action logged, signed, immutable
Sandbox destroyed
No state persists beyond declared memory

Sovereignty by default

Local models run entirely on your hardware. No mandatory API calls to foreign services. Works completely air-gapped.

Compliance-ready audit trail

Every agent run produces an Ed25519-signed, immutable log. Export to your SIEM or compliance platform. GDPR Article 5(1)(f) compatible.

Breach containment built in

A compromised agent cannot escape its sandbox. EuroGuard contains the damage to exactly the declared capability scope.

Open agent API

Build agents in any language that compiles to WASM. Standard tool-calling interface. EuroIPC for agent-to-agent pipelines.

NIS2 & AI Act aligned

Capability isolation, audit logging and human oversight built in — not bolted on. Designed for regulated industries from day one.

Sectors that cannot use cloud AI

Government, healthcare, defence, critical infrastructure — EuroAgent is built for organisations where data cannot leave the building.

Zero Trust by construction

Built on Zero Trust
for AI agents.

Zero Trust is now codified guidance — NIST SP 800-207, the NSA's implementation guides, OWASP's least agency, and a US federal mandate by 2027 — and it is being extended to autonomous AI agents. EuroOS makes its core controls native kernel primitives, designed for breach from the first line of code.

Least agency, enforced at the syscall boundary
Identity-based isolation, not a network perimeter
Hardware-rooted credentials (TPM) and attestation
Immutable, hash-chained audit of every action
Industry validation

The industry is
converging on this.

Sandboxing AI agents, isolating them, and putting guardrails around the tools and credentials they touch is becoming the industry consensus. Red Hat, one of the largest open-source platform vendors, is now publicly building exactly these defenses for agentic AI.

The industry view

Red Hat's guidance calls for defense-in-depth around AI agents: per-agent kernel isolation (OpenShift sandboxed containers), policy-enforced tool and network access (OpenShell), and short-lived, scoped credentials injected at the boundary, so even a compromised agent has nothing to exfiltrate. It even points toward sovereignty.

EuroOS goes one layer deeper

We reached the same conclusion, then built it into the OS. Instead of layering governance on top of Linux and Kubernetes, EuroOS makes the capability the kernel's native primitive: the operating system itself is the sandbox. Agents run capability-isolated in WASM, every tool call is Ed25519-signed and audited, and credentials stay in a capability-gated vault, fully offline and EU-sovereign.

When a vendor of Red Hat's scale builds the same guardrails, it validates the direction. EuroOS was designed around it from the first line of kernel code.

Comparison

Unlike anything
Europe has today.

Every layer built from scratch. That changes what the platform can promise — and prove.

CapabilityEuroOSLinux distributionsWindows / macOSUS cloud AI agents
From-scratch kernel (no Linux/BSD)
Fully auditable by EU governments (EUPL-1.2)Partial
AI agents isolated at kernel level
Agent runs fully offline, no cloud needed
Immutable, signed agent audit trailPartial
Zero telemetry (verifiable)Partial
Runs on embedded / OT hardware (75 MB RAM)Partial
Own TLS stack with European trust store
Governed and hosted in EuropeVaries

A EuroOS ✓ means the capability is built and verified in the kernel today, or is an architectural property of the from-scratch stack. The desktop is shipping alpha; the agent-runtime, OT and sovereign-compute products that package these capabilities are on the roadmap (Horizon E). See the technical docs for exactly what runs today.

NIS2 Article 21 aligned by design

NIS2 for your operational technology.
Without replacing your systems.

EuroOS OT runs on existing hardware. 75 MB RAM. No external dependencies. No telemetry. EuroGuard watches every process. Every network connection is explicitly approved. Every incident logged. Roadmap profile

Memory footprint75–95 MB
Boot time (QEMU / hardware)~1.6 s
Active processes (full desktop)27–29
External dependencies0
Telemetry events0
Verify before executeOnly Ed25519-signed firmware and software runs. A tampered binary is refused before it touches memory.
Own network stack — default denyEvery outbound connection must be explicitly permitted. The firewall is in the kernel, not a userspace app that can be bypassed.
Audit log — NIS2 readyEvery process action, network event and security decision is logged. Signed, immutable, exportable to your SOC or compliance platform.
No hardware replacementRuns on x86-64 hardware from 2010 onwards. Deploy as a gateway or edge node alongside existing OT systems.
Get involved

A sovereign computing stack
Europe has been waiting for.

From kernel to AI agent — auditable, sovereign, built in Rust. If you are building for regulated industries, critical infrastructure, or simply believe Europe deserves its own computing foundation: we want to hear from you.