Vaicoa · Own your platform.
Production on your own servers. Agents work on its twins.
Vaicoa is a small platform you run in your own cloud project: one Go binary with one SQLite file. It creates your servers, keeps them on signed releases, backs itself up, and makes short-lived copies of an environment, twins, where people and coding agents work without touching production.
What you get
Eight benefits, each with its proof
Every claim on this page carries its status. Built runs end to end in a lab. In progress is partly built or partly proven. Planned is designed, not built.
Small footprint
One Go binary and one SQLite file on a 1 GB server, in your own cloud project.
50–62 MB of memory in normal operation (lab, 23 Oct 2026). The hub’s cpx02 costs €7.13 a month incl. VAT (Hetzner’s API, 23 Oct 2026).
Twins
A copy of an environment for every change, with seed data by default, reached only through the hub.
Ready in 44 s on Hetzner; its address out of the hub’s firewall 7 s after delete (20–21 Oct 2026).
Agents with limits
An agent principal with a ceiling the hub enforces, approvals by a person, sandboxed agent servers, MCP tools.
An OpenShell sandbox ready in 78 s, €0.17 a run (lab, 23 Oct 2026). Approvals from the phone are planned.
Works with your agent
Claude Code, OpenCode or any MCP client, each with its own key, never yours.
Through ownpaas-mcp or vai --agent today. Hosted runs (Claude Code in the cloud) are planned.
Built for production
Signed releases, staged updates, continuous backup, recovery tests, an audit kept 400 days, memory limits.
Built and drilled on Hetzner; open: the release signing keys’ ceremony and the conformity with the Cyber Resilience Act.
Platform independence
A Hetzner-compatible API, vaicoa-iaas, with providers as Go modules.
The KubeVirt provider is built. The Hetzner passthrough and an OpenStack provider are planned next.
From local to cloud
The same hub on your Mac in ownpaas and on Hetzner Cloud.
On the Mac: a hub in 13 s, a twin in 22 s (ownpaas lab, as of 23 Oct 2026).
From Podman to Kubernetes
Start with Podman hosts, grow into k3s clusters.
k3s is built in the agent; the hub’s alpha still refuses it.
How it works
Four steps, all in your project
The hub, its database and your cloud tokens stay in your cloud project. Agents get no cloud token, no device key and no production access.
- 1
vai up
puts the hub on a small server in your project and pairs your device as its owner.
- 2
vai host create web-1
The hub creates Podman hosts in that project. Each host’s agent connects out to the hub and keeps the release and settings the hub holds for it; no host listens for the hub.
- 3
vai twin create
A twin is a copy of an environment in a second project, with seed data by default, reached through the hub with a short-lived grant.
- 4
vai --agent claude-code
Coding agents use the hub’s tools with their own key: they work on twins and ask a person before production changes.
One row per environment
The Twin Row: the environment’s state now, its twins behind it, an agent at work, what waits for your approval, and thirty days of production. The hub’s portal draws it; the Vaicoa app’s version is planned.
Measured
Numbers from real runs
Each figure comes from a lab record, with the date of the run. A lab run is not production.
- 79 of 80The stage-2 gate on Hetzner Cloud: rows passedHetzner Cloud, 20–21 Oct 2026
- 44 s / 11 s / 7 sA twin ready / its address in the hub’s firewall / out after deleteHetzner Cloud, 20–21 Oct 2026
- 64 s / 75 sA recovery test on Hetzner, start to server deletedHetzner Cloud, 20–21 Oct 2026
- 50–62 MB / 256 MBThe hub’s memory in normal operation / at a 1 GiB twin download (limit 560 MB)lab run G1, 23 Oct 2026
- 78 s / €0.17An agent sandbox (OpenShell) ready on a new server / one lab run’s costHetzner Cloud lab, 23 Oct 2026
- 13 s / 22 sOn the Mac (ownpaas lab): a hub created / a twin readyownpaas lab, as of 23 Oct 2026
The family
One brand, three companions
The Vaicoa app
The former Cloud Viewer: your Hetzner servers’ costs, logs, metrics, uptime checks and alerts on your phone. Pairing it with a hub, for the Twin Row, approvals with Face ID and push notices, is Planned
ownpaas, by Vaicoa
The cloud on your Mac: a local lab that answers the Hetzner Cloud API and runs real virtual machines, so a hub and its twins can be rehearsed before a server is rented. ownpaas.com
Vaicoa Academy
Courses with hands-on labs: video series, self-paced labs and cohorts, about building and running software on your own servers with coding agents. Under this name it is Planned
Open source
Apache-2.0, public from the launch
The hub, the agent on each server and the vai command are open source under Apache-2.0. The repositories go public at the launch, planned for February 2027, under github.com/vaicoa. The software costs nothing; you pay your cloud provider for the servers. A business plan with SSO and compliance evidence, and support, are planned, with no price yet. Pricing →