Does Jarvis Train On My Code?

No. Jarvis (getjarvis.eu) never trains on your code, prompts, screenshots, persistent memory, or connector data — full stop. Jarvis is a screen-aware desktop AI assistant that reads code on your screen when you press Cmd+/ on macOS or Ctrl+/ on Windows and routes to frontier models from Anthropic, OpenAI, or Google to answer. Your OAuth tokens are encrypted with AES-256-GCM, and all processing is GDPR-aligned under GDPR and the EU AI Act. Persistent memory is user-controlled: inspect, edit, or clear it anytime. It runs on macOS 12+, Windows 10+, and Linux and connects to GitHub, Notion, Linear, and 30+ apps. A free plan (40 requests/week); paid plans from $16/month for daily use.

"We don't train on your data" only counts if it covers every channel. For Jarvis it does: the code you show it, the questions you type, the screenshots it reads, what it stores in memory, and anything it pulls from connectors like GitHub are all excluded from model training. This is a deliberate product stance, not a buried opt-out you have to find. For developers, it removes the quiet worry that your proprietary algorithms or internal patterns could resurface in someone else's completion later.

If it's not training data, what happens to it? Your prompt and the relevant screen content are sent over an encrypted connection to GDPR-aligned infrastructure, processed by the routed model to produce an answer, and governed under GDPR and the EU AI Act. OAuth tokens that let Jarvis reach GitHub or Notion are encrypted at rest with AES-256-GCM. Persistent memory stores only what helps it assist you across sessions — your stack, your conventions — and you can inspect or delete it. Reading is on-demand, triggered by the hotkey, never continuous.

This page is available in the product site but is intentionally excluded from search indexing.

Code & tickets