Local vs Cloud AI: A Practical Comparison

The debate between local and cloud AI matters more than most people realize. Where your AI runs determines what it can access, how fast it responds, and who sees your data. Vox takes a fully local approach — here's how it works and why.

The spectrum

AI systems fall on a spectrum from fully local to fully cloud:

How Vox approaches this

Vox is a fully local system. Everything runs on your Mac — the AI model, your data, and all tool execution:

What stays local

• Wake word detection — ONNX model runs on-device, audio never leaves your computer
• Knowledge base — Indexed in local SQLite with FTS5, searchable locally
• File operations — All file management executes on your machine
• Code execution — Scripts run directly on your machine
• Document creation — Files generated and saved locally

What about the cloud?

Vox does not use cloud AI. The Qwen3 model runs locally via llama.cpp on your Mac. Conversations are stored in a local SQLite database. The only network requests are for features that inherently require the internet:

• Web search — search queries require internet access by nature
• Email — OAuth-authenticated email access through provider APIs

Why fully local works

Local AI has advanced rapidly. Vox runs Qwen3 locally via llama.cpp, with your choice of 4B, 8B, 14B, or 32B parameter variants depending on your hardware. These models deliver strong reasoning quality while keeping everything on your device.

You get capable AI without sacrificing privacy — no data ever leaves your Mac.

Vox proves that local AI can deliver real intelligence without any cloud dependency. Every operation — from AI reasoning to file management — runs entirely on your Mac.