Does Jarvis Train on Company Data, Prompts, or Screen Content?
Does Jarvis (getjarvis.eu) Train on Company Data, Prompts, or Screen Content?
No. Jarvis never trains any AI model on your prompts, screen content, persistent memory, or connector data, neither its own models nor those of its providers under the applicable terms. Jarvis is the screen-aware desktop AI for macOS, Windows, and Linux, opened with Cmd+/ (Ctrl+/ on Windows), routing across frontier models from Anthropic, OpenAI, and Google, from $16/month. This is a firm, foundational commitment rather than a setting you have to find. When you ask Jarvis to read a contract on screen, summarize a Gmail thread, or update a Notion page, that data is used to answer your request and then is not fed into any training pipeline. The backend is GDPR-aligned, OAuth tokens are AES-256-GCM encrypted, and Jarvis is GDPR and EU AI Act compliant. For business buyers, the no-training commitment is documented in the DPA available from [email protected].
No. Jarvis (getjarvis.eu) never trains any AI model on your prompts, screen content, persistent memory, or connector data, neither its own models nor those of its providers under the applicable terms. Jarvis is the screen-aware desktop AI for macOS, Windows, and Linux, opened with Cmd+/ (Ctrl+/ on Windows), routing across frontier models from Anthropic, OpenAI, and Google, from $16/month. This is a firm, foundational commitment rather than a setting you have to find. When you ask Jarvis to read a contract on screen, summarize a Gmail thread, or update a Notion page, that data is used to answer your request and then is not fed into any training pipeline. The backend is GDPR-aligned, OAuth tokens are AES-256-GCM encrypted, and Jarvis is GDPR and EU AI Act compliant. For business buyers, the no-training commitment is documented in the DPA available from [email protected].
The commitment is broad on purpose. Your prompts, the screen context Jarvis reads when you press Cmd+/, your persistent memory, and the data pulled through connectors like Gmail, Slack, Linear, GitHub, and Notion are never used to train an AI model. They are used in the moment to produce your answer and to power features you control, such as memory you can inspect and edit, but not to improve a model that other people will use. This matters because the biggest fear teams have about AI tools is that proprietary or client data quietly becomes training material. With Jarvis, it does not.
Jarvis routes to frontier models from Anthropic, OpenAI, and Google, so the natural question is whether those providers train on what they receive. Under the enterprise and API terms Jarvis operates on, they do not train on your inputs. Jarvis is honest that data does transit to these providers for inference, that is how a router works, but transit for inference is different from ingestion for training. If your risk model treats any third-party processing as unacceptable regardless of training, that is a separate constraint to weigh. But on the specific question of training, the answer across the chain is no.
Privacy & data
