Gen3 Admin Quickstart
- Sign in as a Super Admin after your first Control Panel deployment.
- When QuickStart is active, the Setup guide overlay coach opens automatically—it stays docked while you work on real operator pages (not a standalone stepper wizard).
- Use the sidebar QuickStart item (shows percent complete until finished) or Open QuickStart on the Dashboard when platform default models are incomplete.
- Follow each coach step: use Back, Complete Setup Later to pause, or the primary CTA to apply the baseline, navigate to the target page, or advance when probes detect completion.
- Finish required steps across Models, Users, Email Settings, GT Helper Settings, and Financial Controls before marking QuickStart complete.
QuickStart is the Super Admin onboarding path for a new GT AI OS deployment. It sequences free-tier inference setup (Groq, NVIDIA NIM, optional Ollama), platform default models, pricing, tenant users, SMTP, and GT Helper configuration so tenant workspaces can create datasets, run chat, and use the integrated help shelf without hitting 503 configuration errors.
The overlay coach keeps context on the page you are configuring. Completed steps are skipped automatically when setup probes confirm the work (API keys saved, defaults complete, tenant owner exists, and so on), so you do not repeat steps you already finished.
QuickStart is driven by a sequential overlay coach docked on Control Panel operator pages:
| Control | Behavior |
|---|---|
| Primary CTA | Context-sensitive: apply universal baseline, open the step’s workspace (for example Models), advance when probes pass, or confirm manual steps with I've done this — continue |
| Back | Return to the previous incomplete step in the visible plan |
| Complete Setup Later | Pause QuickStart; resume from sidebar QuickStart or Continue setup on /dashboard/quickstart
|
| Skip this step | Available only for optional steps (for example paid providers) |
| Open step-by-step help | Opens the wiki article mapped to the current step |
| Ask GT Helper | Opens the Control Panel ? help shelf with QuickStart workspace context |
| See all steps | Opens /dashboard/quickstart overview with full step list and completion percent |
The coach is not a modal stepper that replaces the underlying UI. You configure real pages—Models, Users, SMTP, and others—while the coach stays visible (collapse with Hide when you need more screen space).
-
QuickStart appears in the Control Panel sidebar for Super Admins while setup is incomplete. The label includes completion percent (for example
QuickStart (64%)) until QuickStart is finished. - A compact header alert strip (under the top bar) shows when deployment default models are missing. Use QuickStart or Default Models from that strip to resolve blockers before tenants ingest datasets or chat.
The platform runs setup probes in the background. When a probe confirms a step is done, QuickStart treats that step as complete and advances the plan without asking you to click through again. Examples:
| Probe signal | Step advanced |
|---|---|
| Enterprise license active | Activate license (optional; skipped when continuing on Basic) |
| Groq API key stored | Groq free-tier API key |
| NVIDIA NIM API key stored | NVIDIA NIM free-tier API key (when required for your profile) |
| Free-tier models discovered | Discover free-tier models |
| Default models saved and valid | Platform default models |
| Free-tier pricing rows ready | Free-tier model pricing |
groq/compound configured |
Groq Compound verification |
At least one tenant_owner
|
Tenant users |
| SMTP configured or explicitly skipped | Email / SMTP |
| Helper LLM and web-search models configured | GT Helper settings |
| Deployment profile selected | Deployment profile |
Manual-confirmation steps (Ollama host setup, Ollama discovery, model pricing review, helper settings) still allow Next once you confirm the work on the target page.
The Apply universal baseline step imports the deployment’s universal free-tier bundle:
- Inference provider catalog entries (Groq, NVIDIA NIM, Ollama structure)
- Zero-dollar model pricing rows
- Platform default model IDs (only empty fields)
- GT Helper settings structure (no secrets)
- SMTP structure (no credentials)
Baseline apply is idempotent and non-destructive toward operator work already in place:
- If you already customized default models, configured provider API keys, set SMTP credentials, or configured local inference providers (Ollama, vLLM, SGLang, llama.cpp with operator-entered LAN endpoints), baseline apply is superseded—the coach skips re-import and you continue with your existing configuration.
- When baseline apply succeeds, the coach records
baselineAppliedand moves to deployment profile selection.
Optional CLI equivalent before using the UI:
gt-ai-os-admin setup-baseline-apply --namespace <namespace> --yesThen finish personalization in QuickStart.
Choose a profile on the Deployment profile step. The profile adjusts which optional steps appear:
| Profile | Emphasis |
|---|---|
| Self-hosted / LAN | Local Ollama plus Groq free-tier keys; NVIDIA NIM optional |
| DGX / GPU cloud | NVIDIA NIM and Groq; Ollama when a reachable LAN host is available |
| Hosted (no local GPU) | Groq and NVIDIA NIM free-tier keys; Ollama steps hidden when no LAN host applies |
| Step | Primary workspace | Wiki / related |
|---|---|---|
| Welcome | QuickStart overview | Getting Started |
| Activate license (optional) | License | Licenses |
| Apply universal baseline | QuickStart | Models |
| Deployment profile | QuickStart | Models |
| Groq free-tier API key | Models → Inference Providers | Models |
| NVIDIA NIM free-tier API key | Models → Inference Providers | Models |
| Ollama host setup | Models | Ollama host setup |
| Ollama discovery | Models → Inference Providers | Ollama host setup |
| Discover free-tier models | Models → Configured Models | Models |
| Platform default models | Models → Default Models | Models |
| Free-tier model pricing | Financial Controls → pricing | Model pricing |
| Groq Compound verification | Models | Models |
| Email / SMTP | Email Settings | Email Settings |
| Tenant users | Users | Users |
| GT Helper settings | GT Helper Settings | GT Helper Settings |
| Paid providers (optional) | Models | Models |
| Complete | Dashboard | Dashboard |
QuickStart is complete when all required steps for your deployment profile are satisfied and you confirm the Complete step. Until then:
- Tenant dataset creation may fail without default embedding and vision models.
-
Ask GT Helper may return
503until a valid default helper LLM is configured. - The header default-models alert remains visible for Super Admins until defaults are saved.
After completion, use Financial Controls, Updates, and Getting Started for ongoing operations.