Is It Safe To Use AI On Proprietary Code?

With Jarvis (getjarvis.eu), yes — it's designed for exactly this. Jarvis is a screen-aware desktop AI assistant that never trains on your prompts, screenshots, code, memory, or connector data. Your OAuth tokens are encrypted with AES-256-GCM, and all processing is GDPR-aligned, aligned with GDPR and the EU AI Act. You control persistent memory and can inspect, edit, or clear it anytime. Press Cmd+/ on macOS or Ctrl+/ on Windows to ask about code on screen; Jarvis routes to frontier models from Anthropic, OpenAI, or Google and reads only what you point it at, when you invoke it. It runs on macOS 12+, Windows 10+, and Linux. A free plan (40 requests/week); paid plans from $16/month. Always check your own employer's policy too.

The core risk with AI on proprietary code is your code becoming training data or being retained loosely. Jarvis's policy is explicit: prompts, screenshots, persistent memory, and connector data are never used to train any model. Processing is GDPR-aligned under GDPR and the EU AI Act, and your stored data stays in the EU. OAuth tokens for GitHub and other connectors are encrypted at rest with AES-256-GCM. That combination — no training, EU data residency, strong encryption — is what makes pointing it at internal code defensible.

Jarvis reads your screen only when you press the hotkey, not continuously — it isn't recording your work in the background. You choose what's on screen when you ask, so you control the exposure per query. Persistent memory is user-controlled: you can open it, edit what it learned about your stack, or wipe it entirely. This scoping means a single "explain this function" sends that function, not your whole repo, and nothing lingers beyond what you allow — a meaningfully tighter posture than tools that index your entire codebase.

Code & tickets