A mind map that thinks with you. Turn what you're learning into a visual knowledge map where every node is an AI you can talk to: ask focused questions, generate notes, let AI restructure the map — you review the change before it lands, and undo it in one step — and find ideas later with on-device semantic search.

It's a living visual map rather than a scrolling chat box, so the shape of what you're learning stays in front of you — and nothing is uploaded. The app is useful before you install any model at all: local AI is an upgrade, not a prerequisite. Import your existing Freeplane (.mm) and XMind maps to get started.

Private by design. Your maps stay on your machine, encrypted at rest with two-factor encryption — a password AND a separate key file — so neither one alone can ever unlock your data. Real protection, not just a checkbox.

Your own AI, on your own hardware. The LLM runs through Ollama or LM Studio — on your machine or elsewhere on your network — so nothing you write is uploaded and there are no per-token fees or subscriptions. (Support for cloud providers is planned for a future release.) Note that a confined snap cannot install or start a runtime for you: install Ollama or LM Studio on the host and point the app at it.

• Three more local models to pick from — Muse Glimmer, Qwen 3.8 (27B) and, on LM Studio, Bonsai 27B.
• The app now tells you when a model is being loaded or freed —.
• Undo now takes back an applied 'Extend with AI' from the.
• Cutting a node now removes it from the map right away, with a banner to cancel the cut or paste it onto another node.
• A research note now shows which model wrote it, and keeps saying.
• A model's deeply nested bullet list is now typeset in chat instead of appearing as a grey box of raw LaTeX.
• The 'keep or undo?' bar for AI changes no longer runs off the.
• A map-extend from bonsai-27b keeps the nested nodes it asked for instead of losing them to a stray quote.
• When a prompt needs more context than the graphics card can hold, the app now says the reply is being generated on the CPU instead of leaving you watching a blank screen.
• When a model's reply cannot be read as map changes, Extend.
• A large map no longer stops working with Ollama after the first slow reply — the app remembers what your card can hold on the GPU without refusing prompts it has already answered.
• The node style panel now opens centred on desktop instead of stuck to the bottom of the window.
• The AI settings no longer show a leftover 'Found N results'.
• Dialogs now show a divider under the title or above the.
• The interface text size you pick in Settings is kept between launches instead of resetting to normal.
• AI features no longer reload the model between them, so an AI.
• Formulas in AI answers and notes now come out typeset, not as raw LaTeX in a code box.
• AI notes now show formulas typeset, the way the chat does.
• The node style panel now has its Cancel and Apply buttons at the bottom, like every other dialog in the app.

MindMap Chat is also available on Windows, macOS, Android. Get it at https://mindmapchat.app/?from=snap.