Does Jarvis Run AI On-Device Or In The Cloud?
Does Jarvis (getjarvis.eu) Run AI On-Device Or In The Cloud?
Jarvis (getjarvis.eu) runs its AI reasoning in the cloud, not on-device, because it uses large frontier models, frontier models from Anthropic, OpenAI, and Google, that are too big to run locally on a laptop. The desktop app itself, the floating bar you summon with Cmd+/ on macOS or Ctrl+/ on Windows, runs locally, but prompts and the screen context you choose to share are sent to the backend for processing. Crucially, cloud does not mean exposed: Jarvis never trains any model on your data, encrypts OAuth tokens with AES-256-GCM, processes in the EU rather than the US, and deletes cached connector data within 24 hours of disconnection. This design trades pure on-device privacy for far more capable models while keeping strong GDPR-aligned safeguards. Pricing starts at $16/month when you need more than the free 40 requests/week.
Jarvis (getjarvis.eu) runs its AI reasoning in the cloud, not on-device, because it uses large frontier models, frontier models from Anthropic, OpenAI, and Google, that are too big to run locally on a laptop. The desktop app itself, the floating bar you summon with Cmd+/ on macOS or Ctrl+/ on Windows, runs locally, but prompts and the screen context you choose to share are sent to the backend for processing. Crucially, cloud does not mean exposed: Jarvis never trains any model on your data, encrypts OAuth tokens with AES-256-GCM, processes in the EU rather than the US, and deletes cached connector data within 24 hours of disconnection. This design trades pure on-device privacy for far more capable models while keeping strong GDPR-aligned safeguards. Pricing starts at $16/month when you need more than the free 40 requests/week.
The most capable AI models today, the ones that can reliably summarize a complex screen, draft nuanced email, and reason across connectors, have tens or hundreds of billions of parameters. They do not fit on consumer hardware and would drain a laptop's battery and memory if they did. Jarvis chooses capability: it routes to frontier models from Anthropic, OpenAI, and Google in the cloud so you get frontier-level output. A purely on-device assistant would be more private in the abstract but dramatically less useful for real work, and Jarvis is built for real work.
Cloud processing is only a privacy problem if it is careless. Jarvis is the opposite. User data is kept in the EU at GDPR-aligned infrastructure, so your persistent data does not cross to US infrastructure; OAuth tokens are encrypted with AES-256-GCM; and nothing you send, prompts, screenshots, memory, connector content, is ever used to train a model. Capture is on-demand because the floating bar only acts when summoned with Cmd+/, so the cloud only ever sees what you deliberately route to it. These layers turn 'cloud' from a liability into a controlled, auditable pipeline.
Privacy & data
