A mind map with Ollama
MindMap Chat is a mind map whose AI runs on your own computer through Ollama. Every node is something you can ask about, and nothing you write has to leave your machine.
What Ollama does here
Ollama is a local AI server: it downloads an open model and answers requests on your own hardware. MindMap Chat sends it one thing at a time, chosen by you: a single node with its notes, its parent and its children, and an outline of the map. That keeps each request small, which is what makes a local model fast enough to be useful, and it keeps the answer about the idea you pointed at rather than the whole document.
With Ollama behind it, the app can:
- Chat about one node. Open a chat on any node and ask about it. History is kept per node, and you can save an answer as that node's note.
- Chat about the whole map. Ask how the parts fit together. A map chat never edits the map.
- Extend a branch. The AI proposes new child nodes. Apply adds them as one step, highlighted with Keep and Undo, so one undo takes them all back.
- Research a node. Ask a question under Properties → AI Research and the answer becomes the node's note, after you confirm replacing an existing one.
Every node the AI changed carries a small badge; tap it to see which model changed what, and when. When you compare models, Resend asks the same question again with the model you have chosen now.
Setting it up
On Windows, macOS and Linux the app sets Ollama up for you. Open Settings → AI, choose Ollama as the provider, pick a model and press Save. The app installs Ollama if it is missing, starts it and downloads the model, with a checklist of the steps and a Retry button if one fails. The download continues while you work; you can cancel it, and starting again resumes where it stopped.
If you already run Ollama, keep the default address http://localhost:11434/v1, pick a
model you have pulled, and press Save. The
Ollama setup guide covers a manual install step by step.
Which model to pick
The model list is fixed to models that have been tested against the app, each with the memory it roughly needs. Two good places to start:
- Gemma 3n (e4B),
gemma3n:e4b: the lightest to run, about 6 GB of memory, for a laptop without a large graphics card. - Qwen 2.5 (7B),
qwen2.5:7b: a better balance, a 4.7 GB download that needs about 8 GB of memory.
The larger models on the list, such as Qwen 2.5 (32B), want around 24 GB of graphics memory or 32 GB of system memory. Some of them reason before they answer: slower to start, and you can read their thinking under Reasoning in the chat. A model too big for your memory still works, only slowly. The local AI page has the full table.
One Ollama for every device
A phone or tablet cannot run Ollama itself, and an old laptop may not run it well. Run it once on
the most capable computer you have, make it listen on your network by setting
OLLAMA_HOST=0.0.0.0, and open port 11434 on that computer's firewall. On every other
device, open Settings → AI, choose Ollama and enter that computer's address, such
as 192.168.1.50. Your requests then cross your own network and nothing else.
What stays on your computer
With Ollama as the provider, the AI's requests go only to the Ollama server you set. Your maps, notes and chats are stored in a database on your device, encrypted with your passphrase, and there is no account. Search does not use Ollama at all: it runs on a small model built into the app, so it works with the network off and with no AI set up. The privacy policy lists every connection the app makes.
If one day you want a cloud model for a single question, Claude and OpenAI are there as an option, with your own key, and only after you agree. Ollama stays the default.