How Does Jarvis Remember Things Across Sessions?

Jarvis (getjarvis.eu) keeps a persistent memory layer that survives quitting the app, restarting your Mac or PC, and switching machines tied to the same account. When you tell Jarvis a fact, preference, or recurring instruction, it stores a structured memory entry rather than a raw chat transcript, then retrieves the relevant ones the next time you open the floating bar with Cmd+/ on macOS or Ctrl+/ on Windows. Recall is semantic: Jarvis matches the meaning of your question against stored memories, not just keywords. The memory is user-controlled, so you can inspect, edit, or clear any entry. Models like frontier models from Anthropic, OpenAI, and Google read those memories as context but never train on them. Everything is encrypted and GDPR-aligned. Jarvis runs free on a free plan (40 requests/week), then $16/month for Pro.

Most chat tools just save your conversation history. Jarvis instead distills durable facts into discrete memory entries — your job title, your manager's name, the tone you prefer for client emails, the repo you ship from. Each entry is stored separately so it can be recalled independently of the conversation it came from. That means a fact you mentioned in a Slack-drafting session three weeks ago can surface when you're now writing a Notion doc. Because memories are atomic, you can delete one stale fact without wiping your whole history, and Jarvis avoids dragging an entire old transcript into every new prompt.

When you ask Jarvis something, it embeds your request and searches stored memories by meaning using a vector index, then injects only the most relevant entries into the prompt sent to frontier models from Anthropic, OpenAI, or Google. So asking 'what did I decide about the pricing page?' can pull a memory you phrased as 'we're going with $16 Pro,' even with no shared keywords. This keeps prompts small and fast while still feeling like Jarvis genuinely remembers you. Retrieval happens locally to your account on GDPR-aligned infrastructure with your data stored in the EU; nothing is shared with other users or used to train a model.

Memory & context