[Feature]: Admin defined global LLM catalog for provider.

[atharvamhaske]

Summary

Think of this as persona: an internal senior Admin defines a curated LLM catalog at Helm install time. Those models show up in the OpenEverest UI as a dropdown so users/developers deploy with one click (no raw YAML). Inference stays OpenAI-compatible for end users; they offload config manifests to OpenEverest + KubeAI via provider-kubeai.

Why

KubeAI’s model catalog is install-time. Today the provider only exposes free-text source/profile fields. Catalog models configured at install must also appear in the UI.

Scope

• Add Helm modelCatalog (or equivalent) with curated entries (name, source, engine/version, resource profile, defaults).
• Surface catalog on the Provider CR so UI can bind a select (optionsPath).
• UI: model dropdown from catalog; optional advanced/raw fields.
• Sync still creates a correct KubeAI Model; clients use /openai/v1/... as usual.
• Docs for Admin catalog + user pick flow (changes = Helm upgrade).

Checklist

• Admin can set catalog in Helm (e.g. Ollama + Deepseek).
• Catalog models appear in Everest UI dropdown.
• Instance from catalog needs no hand-edited source YAML.
• Sync → valid KubeAI Model; OpenAI API works.
• Docs cover install-time catalog semantics.

Additional notes

Touch points: topology.yaml (UI), versions.yaml (engine versions ≠ model catalog), charts/provider-kubeai/values.yaml, deploy/kubeai/values-gpu.yaml.

Resources

• https://www.kubeai.org/reference/kubernetes-api/
• https://www.kubeai.org/

[spron-in]

Makes sense. I'm not sure if we can use newly created Presets here for that. Like Preset will define which model is available for which user.
Preset spec:
https://github.com/openeverest/specs/blob/main/specs/006-preset.md

[alokkumardalei-wq]

@atharvamhaske @spron-in I'd love to pick this up!
@spron-in, I agree,using InstancePresets is definitely the way to go. It completely saves us from having to build and maintain a custom catalog system from scratch.
Instead of adding a new modelCatalog array, we could just bundle a few default InstancePreset manifests (like Llama 3, Deepseek, etc.) right into the provider-kubeai Helm chart. The Everest UI will then naturally pick them up for the dropdown.

Let me know if this approach sounds good to you both!