Κυβερνήτης · the steersman
The agentic OS for your business. Your AI operations stay yours.
Connect your repositories, documents and tools. Give an orchestrator agent instructions. Kybernetes OS runs the work in isolated sandboxes on open-weight models, verifies the result in a real browser, and ships a pull request with evidence. Installed on your infrastructure, with the GPUs you choose, and nothing leaves unless you say so.
Installed on premises or in your cloud accounts · open-weight models by default · EU vendor
What it is
One installation. Every agent your company runs, governed in one place.
Agents with a chain of command
An orchestrator agent receives goals and delegates to specialists. You set who may call whom, with budgets, approval gates and working hours per agent.
Work that is verified before you see it
Tests run, a browser opens the result, a review pass checks the diff. Failed attempts are retried with feedback or escalated. Humans see finished work with evidence attached.
Staff, roles and audit built in
Your identity provider signs people in. Owners, admins, agent managers, reviewers and finance each see what their role allows. Every file read, command and model call is logged, append-only.
How it works
A Node in your perimeter. A small vendor hub that never touches your data.
Your Node
- Console, orchestrator, task queue, approvals
- Isolated sandboxes per task with a real browser
- Open-weight models on your GPUs or GPU accounts, spawned when the queue needs them and released when idle, on the schedule you set
- Connectors to email, calendar, chat, tickets, CRM through your own OAuth grants
- Your database, your object storage, your keys
Kybernetes hub
- Issues your licence and signed software releases
- Receives health heartbeats and usage numbers for billing
- Hosts the model store and registries you install from, every file hash-verified
- Support sessions only when you open one, time-limited and recorded
Sovereignty by construction
You decide which models and providers may see your data. The gateway enforces it.
Every model route carries an exposure class. Data sources and repositories carry a label. A request that would send Restricted data to a route above its class is refused, not logged and regretted.
Hash-verified supply chain
Models, container images, charts and packages come only from stores we operate, mirrored across independent operators. Every file's hash ships inside the signed release. Your Node verifies before it loads, and never calls Hugging Face, Docker Hub or a hyperscaler at runtime.
Upgrades that cannot break your data
Backup, drain running tasks, additive schema migration, apply, smoke test, roll back automatically if anything fails. Every version runs against the previous version's database, so rollback always works.
Models you choose
Open weights first. Routed by task, sized to the GPUs you are willing to run.
Prefer a frontier API for the hardest 20 percent of tasks? Enable it per tenant, per agent, with your own keys. It stays off until you turn it on.
Pricing
A licence per installation. Usage on your own GPUs at your own cost.
50 seats · 25 agents · OIDC single sign-on · standard connectors · 8×5 support
- Onboarding €10,000 one-off
- €1.00 per agent-hour, metered
Unlimited seats · 100 agents · SAML and SCIM · SIEM export · confidential-computing lane · 24×7 support
- Onboarding €25,000 one-off
- €0.80 per agent-hour, metered
- DORA and AI Act evidence packs
Air-gapped installation on your hardware · flat per-GPU licence · on-site days · source escrow
- From €120,000 per year
- No metering, no telemetry if you prefer
Contact
Start with a two-week pilot on your own repositories.
We install a Node in your cloud account or on a server you provide, connect one or two repositories, and measure cost per completed task against your current process.