We use lightweight, privacy-friendly analytics to see which features land — never your maps, never your AI keys, never sell your information. You can decline; the app works fine either way.
Ask any question across every map you've made. Get a grounded answer with citations.
Press ⌘/Ctrl + K from anywhere in the app, or click the magnifying glass in the sidebar. The Research panel slides in from the right with an empty question box.
Type a specific question. 'What does Popper say about falsifiability?' beats 'Popper'. The more precise your prompt, the tighter the citations. Press Enter to submit.
Claude streams the response token-by-token. Every claim comes with a citation chip — hover to see the source snippet, click to jump to the exact map element or PDF highlight it's drawn from. Nothing is invented — if the corpus doesn't know, the assistant says so.
For complex questions flip the Deep Research switch in the panel header. The assistant chains multiple searches, explores relationships, and builds a longer, better-structured answer. It takes longer but handles multi-step questions beautifully.
Click 'Save to Memory' to pin the answer — it becomes a research note accessible from the /memory page. Subsequent questions automatically consider your saved notes as context, so your assistant gets smarter the more you use it.
Every paragraph of the answer has a quick-add button. Click it to insert that paragraph as a new map element on your current map, with the citation preserved as a sub-note. Your map grows organically alongside the conversation.