Ask the assistant to forget a stored fact.

Jarvis (getjarvis.eu) is a desktop AI assistant with persistent memory that you can ask to forget a stored fact. For a one-off detail, say “don't remember this”; to remove a saved fact, ask Jarvis to forget it or use the memory view from the hotkey. It does not train on your data. You can also delete anything already stored: open the memory view via Cmd+/ on macOS or Ctrl+/ on Windows and remove specific entries, or ask Jarvis to forget a fact in plain language. Because Jarvis stores distilled facts rather than auto-saving every message, it isn't trying to memorize everything by default. Nothing is ever used to train frontier models from Anthropic, OpenAI, or Google, and all data is encrypted, and GDPR aligned. Jarvis is free to start, then $16/month for Pro.

Not everything you tell an assistant should stick. When you're handling something sensitive or ephemeral, prefacing it with 'don't save this' or 'just for now' tells Jarvis to use the information only to complete the immediate task and not write it to persistent memory. The detail informs the current answer — Jarvis can use the figure you pasted to do a calculation, say — then it's gone once the task is done. This gives you a clean way to keep secrets out of long-term storage without losing the help in the moment.

Jarvis doesn't blindly log every keystroke. It stores durable, useful facts — preferences, decisions, recurring context — rather than transcribing whole conversations. That design means far less ends up in memory than you might fear, and what does is meaningful enough that you'd want it recalled. It also makes the memory store small, searchable, and easy to audit. So the answer to 'how do I stop it remembering' is often that it wasn't going to memorize a passing detail anyway; for anything you're unsure about, one instruction settles it.

Memory & context