Mind: learning path
Mind is Buraaq AI. Complexity stays in the toolchain: you type intent, the planner picks backend, precision, and layout. Strata is how Mind survives models that do not fit in VRAM.
Read these in order. Each page is one job.
| Level | Page | You can |
|---|---|---|
| Basic | AI basics | Doctor, inspect, chat, std.ai |
| Intermediate | Serve models | OpenAI-compatible API, status, streaming |
| Intermediate | Strata (heavy models) | GPU → RAM → NVMe residency plans |
| Advanced | Train adapters | LoRA/QLoRA via managed venv |
| Advanced | Deploy AI | Pack, ship preflight, Land --ai |
One sentence
Native AOT is still the default for apps. Mind adds local models, serve, and training without turning your project into a Python notebook. Strata tiers weights when the model is larger than VRAM.
Honest boundaries
Inference engines and the train venv are peer processes. Land never auto-installs GPU drivers. Large weights stay in ~/.buraaq/models, not inside every .bur. Strata Phase 1 orchestrates llama.cpp (and limited vLLM); it is not a native tensor pager yet.
Next: AI basics.