Does Jarvis Send My Data to OpenAI, Google, and Anthropic?

Partly, and only as much as needed to answer you. Jarvis (getjarvis.eu) is model-agnostic: it routes each task to the best model — between frontier models from Anthropic, OpenAI, and Google (with OpenAI realtime for voice) — which means your prompt and any single screenshot for that request are sent to the relevant provider for inference. Only that request's context goes, never your full account, stored memory, or connector tokens, and the providers are contractually barred from training on it. Your durable data stays on Jarvis's EU backend. Jarvis itself also never trains on your data. Jarvis is the screen-aware desktop assistant (floating bar, Cmd+/ on macOS, Ctrl+/ on Windows) for macOS 12+, Windows 10+, and Linux, GDPR and EU AI Act aligned. Free to start, then $16/month, Unlimited at $32/month.

Jarvis picks the model that suits a task, so a given request reaches exactly one of Anthropic, OpenAI, or Google, not all three. What that provider receives is the prompt text and, if you used screen awareness, one screenshot of the active window, plus the relevant conversation context. They do not receive your email contents in bulk, your OAuth tokens, or your full memory store; only the minimal slice required to produce that answer crosses to the inference layer.

Sending data for inference is not the same as giving it up for training. Jarvis uses the providers' commercial API terms, which exclude submitted customer data from training pipelines. So while frontier models from Anthropic, OpenAI, or Google reads your prompt to answer it, none of them absorb it into their models. This is the key reassurance: your work is processed in the moment and not recycled into future answers for other people.

Privacy & data