Why Privacy-First AI Matters More Than You Think
Every time you paste a document into ChatGPT or upload a file to Claude, that data leaves your computer and lands on someone else's server. For personal notes, that's fine. For client contracts, financial records, medical information, or proprietary code? That's a risk worth understanding.
Where your data goes
With most AI tools, every interaction is sent to a cloud server for processing. The provider's terms determine what happens next:
- Can they train on your data? (Many can, by default)
- How long is it retained? (Often unspecified)
- Who has access? (Employees, subprocessors, partners)
- Where is it stored? (May cross jurisdictions)
Vox's privacy architecture
Wake word detection
ONNX model runs on-device. Audio is processed locally and discarded. Nothing is streamed to any server.
Knowledge base
Files are indexed into a local SQLite database on your machine. Excerpts are passed to the local AI model only when you ask.
File operations
All file reads, writes, moves, and searches happen locally. File content is only shared when you explicitly ask Vox to use it.
Code execution
Scripts run directly on your machine. Code never leaves your Mac.
Everything stays on your Mac
Unlike cloud AI tools, Vox runs entirely on your machine. There is no server, no cloud model, and no data leaving your device:
- AI reasoning — Qwen3 runs locally via llama.cpp on your Mac, no cloud API calls
- Conversation history — stored in a local SQLite database on your machine
- Email access — OAuth-authenticated, no passwords stored
Everything stays local — your files, knowledge base, code execution, AI reasoning, and conversation history all live on your Mac. Nothing is sent to any cloud server. This is fundamentally different from tools that require you to upload everything.
Why this matters for professionals
- Lawyers — client privilege requires data control
- Healthcare — HIPAA compliance limits data sharing
- Finance — regulatory requirements for data handling
- Developers — proprietary source code needs protection
- Executives — strategic plans and M&A data are sensitive
True local-first privacy
With Vox, the AI model runs on your Mac via llama.cpp — your prompts never leave your device. There is no cloud exposure to minimize because there is no cloud.
Every operation — file access, knowledge indexing, code execution, voice detection, and AI inference — happens entirely on your machine. The model sees only what you ask it to see, and nothing ever leaves your Mac.
“Computer, search my local knowledge base for the contract terms with Acme Corp.”
The search happens locally. Only the relevant results are passed to the local AI model for reasoning — not the entire contract.
Privacy-first AI means keeping everything on your machine — your data, your conversations, and the AI model itself. No cloud, no compromise. That's what Vox was built for.
Put Vox to work on your computer.
Download Vox for Mac and start with the local setup flow.
macOS · Apple Silicon & Intel