Improving AI into a learning asistant

[AnujaKalahara99]

I noticed the repo already has two AI pieces:

aidebugger.js — a context-aware chat UI that POSTs the current project to a backend (/analyze).
aiwidget.js — an AI music generator widget (currently commented out in the palette) that calls an external API client-side using env.GROQ_API_KEY.
Wanted to ask about long-term plans for AI integration and raise a few concrete questions/concerns:

Roadmap

Is there a planned roadmap for promoting these to first-class features (AIDebugger → Learning Assistant agent, AIWidget → music generation), or will AI remain experimental?
Is there a dedicated backend repo or deployment we should target when contributing?
Models & infra

Which LLMs are intended/approved for production use (local LLMs vs cloud models like Gemini)?
Any constraints on latency/cost or preference for RAG / vector DBs?