-
Notifications
You must be signed in to change notification settings - Fork 0
Gen3 Self Hosted Installation
Host-level install for GT AI OS on a single machine using the Quick Installer .deb and the interactive operator wizard. Commands are the same on Ubuntu and NVIDIA DGX; the wizard only asks you to confirm detected host type and CPU architecture.
For post-install Control Panel setup and handoff to GT AI OS Instructions and GT Helper, continue to Self-Hosted Control Panel setup.
- Ubuntu 24.04 (x86_64) or DGX OS 7 (ARM64): 8+ CPU cores, 16 GB RAM, 100 GB disk recommended
- Outbound HTTPS to GitHub Releases and
ghcr.io/gt-edge-ai(or your approved mirror) - A static LAN IP you will use for HTTPS access on ports 3001 (Control Panel) and 3002 (tenant app)
- StorageClass suitable for CloudNativePG (for example Longhorn with immediate volume binding)
v3.0.2+ uses registry-backed images only. Public releases on GT-Edge-AI/GT-AI-OS do not require a GitHub PAT when release assets and GHCR packages are publicly reachable.
Replace v3.0.3 with your target release tag. The .deb filename uses semver without the v prefix (for example 3.0.3 for tag v3.0.3).
curl -fsSL -o /tmp/gt-ai-os.deb \
https://github.com/GT-Edge-AI/GT-AI-OS/releases/download/v3.0.3/GT-AI-OS-Quick-Installer_3.0.3_all.deb
sudo apt install -y /tmp/gt-ai-os.deb
sudo -E gt-ai-os-operatorAt the operator menu choose 1) Install, then 1) Interactive (recommended).
| Prompt | Recommended answer |
|---|---|
| What do you want to do? | 1 Install |
| Install — choose style | 1 Interactive |
| GitHub personal access token |
Skip if anonymous download from GT-Edge-AI/GT-AI-OS works; otherwise paste a fine-grained PAT with Contents: Read on that repo |
| Prompt | Default | Recommended |
|---|---|---|
| Detected host type — use this? | Y | Y (Ubuntu or DGX as detected) |
| Select host type (if not detected) | [1] | 1 Ubuntu or 2 DGX OS 7 |
| Detected architecture — use this package? | Y |
Y (linux-amd64 or linux-arm64) |
| Kubernetes namespace | gt-ai-os-prod |
Enter to keep default |
| GitHub authentication (if prompted) | [1] PAT / [2] gh login |
Usually skipped on public releases; if shown, 1 or 2 |
| Choose a release | [1] = latest | 1 or Enter |
| Prompt | Default | Recommended |
|---|---|---|
| Control Panel — access model | [1] | 1 LAN only |
| Tenant App — access model | [1] | 1 LAN only |
| Control Panel LAN host | auto-detected IP | Your static LAN IP (no https://, no :3001) |
| Tenant App LAN host | same as Control Panel | Enter (same IP) |
NAT & DNS or Cloudflare choices add FQDN, profile, and API token prompts. Use those only when you need public hostnames or tunnels; the wizard walks through each field after you select 2 or 3.
| Prompt | When shown | Recommended |
|---|---|---|
| How should GT AI OS use Kubernetes? | Always (unless env preset) | 1 Auto-detect |
| Fresh install vs Resume | RKE2 already running, one node | 1 Fresh on clean host; 2 Resume to keep cluster |
| Wipe and install fresh vs Abort | RKE2 artifacts but API down | 1 Wipe for broken reinstall |
| Proceed with namespace-only install? | Existing multi-node cluster | Y only if you intend namespace-only |
On a clean host, auto-detect bootstraps a single-node RKE2 cluster without extra prompts.
| Prompt | Recommended |
|---|---|
Ready to install? (y/N)
|
y |
Install then runs packages, Helm, and validation. gt-ai-os-admin is installed to /usr/local/bin as part of this flow.
To skip LAN host prompts:
export GT_AI_OS_CONTROL_PANEL_LOCAL_EDGE_HOST="<your-lan-ip>"
export GT_AI_OS_TENANT_APP_LOCAL_EDGE_HOST="<your-lan-ip>"| App | URL | First login |
|---|---|---|
| Control Panel | https://<lan-ip>:3001/login |
gtadmin@test.com / Test@123
|
| Tenant app | https://<lan-ip>:3002/login |
gtadmin@test.com / Test@123
|
Accept the browser self-signed certificate warning for local HTTPS.
sudo env GT_AI_OS_ADMIN_CONFIG_DIR=/var/lib/gt-ai-os/admin \
KUBECONFIG=/etc/rancher/rke2/rke2.yaml \
PATH="/var/lib/rancher/rke2/bin:/usr/local/bin:$PATH" \
gt-ai-os-admin validate --namespace gt-ai-os-prod
sudo env GT_AI_OS_ADMIN_CONFIG_DIR=/var/lib/gt-ai-os/admin \
KUBECONFIG=/etc/rancher/rke2/rke2.yaml \
PATH="/var/lib/rancher/rke2/bin:/usr/local/bin:$PATH" \
gt-ai-os-admin report --namespace gt-ai-os-prodReplace gt-ai-os-prod if you chose another namespace.
| Symptom | What to check |
|---|---|
| ImagePullBackOff on app pods | Cluster must pull ghcr.io/gt-edge-ai/gt-ai-os-*:<tag>; packages must be public or ghcr-pull configured |
| Release download 401/403 | Register GitHub auth once: gt-ai-os-admin auth register (or paste PAT when the wizard prompts) |
| Database pods pending | StorageClass and volume binding for CloudNativePG |
| Wrong URLs after install | Re-run gt-ai-os-admin report --namespace <ns>; confirm LAN IP matches install prompts |
Report install runbook gaps on GitHub Issues.
Continue to Self-Hosted Control Panel setup for passwords, license, models, and enabling GT Helper before inviting tenant users.