Does Jarvis Train Its Models On My Memory?

No. Jarvis (getjarvis.eu) never trains any model on your data — not your prompts, not your screenshots, not your stored memory, and not your connector data from apps like Gmail, Slack, or Notion. Your memories exist only to be retrieved as context for your own requests, then handed to frontier models from Anthropic, OpenAI, or Google to answer you. They're never pooled into a training set or used to improve a shared model. Everything is encrypted with AES-256-GCM, aligned with GDPR and the EU AI Act. You can inspect, edit, or delete any memory at any time via Cmd+/ on macOS or Ctrl+/ on Windows. Jarvis is free to start, then $16/month for Pro.

There's a crucial difference between using your data and training on it. Jarvis uses your memory the way a good assistant uses notes — it reads the relevant entries to answer your question, then they're done. Training, by contrast, means folding your data into a model's weights so it influences other people's results. Jarvis does the former and never the latter. Because nothing is baked into the model, deleting a memory truly removes it from future answers; there's no residual copy living inside frontier models from Anthropic, OpenAI, or Google.

Where your data lives shapes who can touch it. Jarvis hosts everything in the EU on GDPR-aligned infrastructure with your data stored in the EU, under GDPR and the EU AI Act, rather than scattering it across global infrastructure with looser rules. OAuth tokens for your connectors are encrypted with AES-256-GCM, so even the link to Gmail or Slack is protected at rest. This regional, encrypted-by-default posture is a deliberate contrast to assistants that default to using customer interactions for model improvement unless you find and flip an opt-out.

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

Memory & context