A mind map with LM Studio
MindMap Chat can use LM Studio as its AI. Each node of your map becomes something you can ask about, and the model answering runs on hardware you own.
Why LM Studio
LM Studio is a desktop app that downloads open models and serves them on your computer. If you already use it to try models, MindMap Chat can use the same installation: the models you have downloaded, the server you already start. It is one of the two local runtimes the app supports, next to Ollama, and the choice between them is yours; the local AI page compares the two.
One model on the list runs only through LM Studio: Bonsai (27B), a 27-billion parameter model quantised to 1 bit, so it fits in about 8 GB of memory at some cost in reliability, and reasons before it answers.
What the AI does in the map
- Node chat. A conversation about one node. The AI sees that node, its notes, its parent and children and the map's outline, and nothing else. Save an answer as the node's note.
- Map chat. A conversation about the whole map that never changes it.
- Extend. The AI proposes new nodes under a node, or across the map. Apply them as one step and take them back with one undo.
- AI Research. Ask a question about a node and the answer becomes its note.
Nodes the AI changed show a badge that says which model changed them and when, so a map built over months with several models still tells you where each idea came from.
Setting it up
On Windows and macOS, choose LM Studio in Settings → AI, pick a model and press Save: the app installs LM Studio if needed, starts its server and downloads the model, with a checklist and Retry. On Linux, install LM Studio yourself from its AppImage first.
By hand, on any desktop:
- Install LM Studio and open it once: that installs its
lmscommand, which the app needs. - Download a model, for example
lms get qwen/qwen2.5-7b -y. - Start the server:
lms server start -p 1234. - In Settings → AI, choose LM Studio, keep
http://localhost:1234/v1, pick that model and press Save.
The first reply can take a minute while LM Studio loads the model into memory; later replies are quicker. The LM Studio setup guide has the details for each platform and a troubleshooting table.
Downloads you can leave running
Models are large, from about 4 GB to over 20 GB. The download icon next to the model list shows every model: downloading, failed, installed or available. With LM Studio, Detach stops watching a download while LM Studio carries on with it, and Reattach picks it up again. A download marked Stalled has made no progress for a while; it has not failed.
One server for the whole house
Start LM Studio's server on your strongest computer with
lms server start --bind 0.0.0.0 -p 1234, download the models there and open port 1234
on its firewall. On a laptop, a phone or a tablet, choose LM Studio in Settings → AI
and enter that computer's address. A remote LM Studio cannot download models for you, so add them
on the server itself.
Private by default
Your maps, notes and chats live in an encrypted database on your device, and there is no account. With LM Studio as the provider the AI's requests go only to the server you set, and search runs on a model inside the app, offline. Read the privacy policy for every connection the app can make.