Install S7 SkyQUBi
Release status (as of 2026-04-14): S7 SkyQUBi is in the pre-launch testing window. Full public launch is July 7, 2026 · 7:00 AM Central (GO-LIVE Release 7). The install instructions below are stable for the current testing deployment; the full production deployment with the immutable- fork rebuild architecture lands with the GO-LIVE ceremony. See README.md for the full release status and path to 2026-07-07.
S7 runs as a set of containers inside a single Podman pod on your own Linux machine. No cloud dependency. No external telemetry. Everything you install here stays on your hardware.
Prerequisites
- Linux (Fedora 40+ recommended; any systemd-based distro with
podmanavailable works) - Podman 4.x or newer —
sudo dnf install podmanor equivalent - ~30 GB free disk for models + database
- ~16 GB RAM minimum (32 GB recommended for the full 7-witness set)
- A GPU is optional — the 3+1 lite witness set runs on CPU-only hardware; the 7+1 full set benefits from any modern GPU
- Outbound internet access for initial model download only — the system can run fully offline afterward
First install
Clone the public repository:
git clone https://github.com/skycair-code/SkyQUBi-public.git
cd SkyQUBi-public
Start the pod:
./start-pod.sh
This brings up:
- PostgreSQL (port 57080, bound to
127.0.0.1only) - Qdrant vector database (port 57090)
- Ollama model host (port 57086)
- The SkyQUBi engine container that connects them
The start-pod.sh script is idempotent — running it twice is safe. See skyqubi-pod.yaml for the full pod definition.
Downloading the witnesses
First run will pull the three lite-witness models (~6 GB total):
qwen2.5:3bdeepseek-coder:1.3bqwen3:0.6b
Plus the embedding model all-minilm:latest (~23 MB).
For the full 7+1 witness set, see the advanced install section in DEPLOY.md (larger download, ~60 GB).
Verifying the install
After the pod is up, run the lifecycle test:
./s7-lifecycle-test.sh
You should see 40/40 PASS. If anything fails, check:
- Pod status:
podman pod ps - Container health:
podman ps --all - Ollama reachable:
curl http://127.0.0.1:57086/api/tags - Logs:
podman logs s7-skyqubi-engine
Ports you should know
All S7 services bind to 127.0.0.1 by default — nothing is exposed to your LAN without explicit configuration.
| Port | Service |
|---|---|
| 57080 | PostgreSQL |
| 57086 | Ollama |
| 57088 | Public chat API |
| 57090 | Qdrant |
If you want to reach the system from another machine on your network, configure a reverse proxy (Caddy, nginx, Traefik) — do not rebind the services themselves. Exposing a database or an inference engine directly to a network is an invitation to trouble.
Uninstall
podman pod stop s7-skyqubi
podman pod rm s7-skyqubi
That removes everything. The downloaded models live under ~/.ollama/ and can be deleted separately if you want to reclaim disk.
Getting help
- Open an issue on GitHub
- Email omegaanswers@123tech.net
Keep in mind: the install is currently in “Core in Development” status before the July 7, 2026 public launch. Expect rough edges until that date. File issues freely.