Suggestion: Consider Rust + Python with stdin/stdout or WebSocket for clean offline architecture

[yaro123456]

Hey, really cool project — love the idea of a fully offline AI assistant.

That said, I noticed the current stack relies on Tauri + llama.cpp + frontend WebView. It works, but it introduces a lot of complexity for cross-platform builds and model management.

Wanted to suggest an alternative architecture that might simplify things and actually improve performance and dev experience:

Proposed Stack:

Rust binary: just handles process orchestration (start model, watch, reload, etc.)

Python subprocess: runs inference using transformers, GGUF, LoRA, or whatever

Rust ↔ Python via stdin/stdout (or WebSocket if you need async streaming)

Optional UI: browser frontend or TUI, depending on use case

Why this might be better:

Removes Tauri complexity — no weird UI layer or Electron-style overhead

You can use real Mistral + LoRA via Python libs

Easier to debug each layer independently

Works better cross-platform (even on old machines)

Cleaner separation of logic: UI ↔ Orchestrator ↔ Inference

Let me know if this is interesting! Would be fun to PR a minimal version of this setup if you're open to it.

Cheers

[david-menloai]

hi @yaro123456
thanks a lot for your suggestion.
cc @louis-menlo here to follow up