My local LLM turns any file into a mind map and it’s actually brilliant

I went deep into electronics a while back and wanted to get back to tinkering. The problem was that, although I already had a good idea of what I needed and what to do, I’d forgotten many of the specifics. To get back up to speed, I’d have to reread manuals, datasheets, and online guides to remind myself how certain components work or which protocols to follow. But I’d already learned all this before and didn’t want to flip through pages again just to find the bits of information I needed. What I needed was something visual I could use as a quick reference to refresh my knowledge. So, I decided to create mind maps.

To make it easier on myself, I went ahead and used AI to analyze my files and generate mind maps I could reference for my upcoming projects. It took some trial and error, but once I dialed in my prompts and workflow, the process turned out to be surprisingly easy.

My system for making mind maps using local AI

My two-step method

Using Obsidian and LM Studio to create mind maps — Screenshot by Jayric Maning –no attributions required

The core idea behind my setup centers on using Obsidian, since it already includes the tools and plugins needed to create mind maps effortlessly. I use two plugins for this setup: one for analyzing my files and generating Markdown formatted specifically for mind mapping, and another for transforming that Markdown into a mind map. You can use any combination of plugins, tools, or services as long as they achieve the same goal. I just prefer Obsidian since many of my reference notes are web clips that I’ve turned into an Obsidian library.

Outside of Obsidian, I use LM Studio to host the model that powers everything. While there are many local LLMs you can try right now, I find that most 8b (8 billion parameters) models, like Gemma and Dolphin, work well if you’re concerned about your computer’s performance. Of course, you can always expect better results from larger models. Smaller models run faster but sometimes miss nuances, while larger models provide better analysis at the cost of processing time.

How I set up my system

Using Obsidian and LM Studio

Install both Mindmap Nextgen and Copilot in Obsidian — Screenshot by Jayric Maning –no attributions required

To set this up, I first installed two plugins in Obsidian. The first one is Copilot by Logan Yang, which connects with my local LLM and brings AI capabilities directly into my notes. Then I installed Mindmap NextGen by James Tindal. This takes the Markdown text and renders it as an interactive visual.

Launch local LLM using LM Studio — Screenshot by Jayric Maning –no attributions required

Next, I downloaded LM Studio and chose a model that fits my specs. My machine has decent hardware but nothing extreme, so I stuck with models in the 8b to 13b parameter range. After installing LM Studio, I opened its settings and launched the local server. Back in Obsidian, I opened the Copilot settings and added my local model by entering the API endpoint provided by LM Studio. I tested the connection, and it worked perfectly. Now, when I open Copilot’s chat window, it communicates directly with my local model.

Creating mind maps using my setup

Turn files into mind maps

Open Copilot in ribbon to access AI sidebar — Screenshot by Jayric Maning –no attributions required

To create a mind map with this setup, I click the Copilot quick-access button on the left ribbon in Obsidian. This opens a sidebar on the right where I can access my AI model and generate the Markdown.

Set up Copilot sidebar to use local model and add note as context — Screenshot by Jayric Maning –no attributions required

Next, I open my note, click the Add Context button, and select Active Note. To specify my AI model, I click the model dropdown at the bottom, choose my loaded Dolphin 3 model, and set it to Chat mode.

Using fine-tuned prompt for mind map creation — Screenshot by Jayric Maning –no attributions required

In the prompt box, I paste my carefully curated prompt and hit Enter. The prompt instructs the AI to analyze main ideas, filter out front matter and unnecessary text, organize the ideas into a hierarchy, and format everything as Markdown suitable for mind mapping. I’ve also used AI to help me create the prompt itself, refining it over time to make sure it captures structure, context, and relevance accurately.

After generating the Markdown, I click the Save as Note button in Copilot. This saves the conversation as a note inside the Copilot folder. Then, I open Mindmap NextGen, point it to that note, and watch it generate an interactive mind map. I can collapse branches, expand sections, and navigate the information visually.

Serial protocol mind map using Mindmap Nextgen — Screenshot by Jayric Maning –no attributions required

The first time I tried this, the output was a mess. The AI produced a flat list with no real structure. After a few iterations, it started generating clean and organized Markdown that made for excellent mind maps.

Online alternative that’s worth considering

In case you don’t like hosting your own setup

Xmind's official web portal — Screenshot by Jayric Maning –no attributions required

While my main workflow uses Obsidian and LM Studio, I also tried Xmind to see how it compares. Xmind offers customizable templates, presentation modes, and built-in AI features that make it appealing if you want to skip the local setup entirely. You can upload your files and let Xmind’s AI generate mind maps automatically without hosting anything on your own machine.

However, prompting here feels a bit more restrictive. With my local LLM approach, I can write highly detailed prompts that specify exactly what to extract, how to organize it, and how to format the hierarchy. With Xmind, prompt length is limited, so you have to keep submitting shorter prompts when doing On-Demand refinements. This works fine if you have a premium subscription; otherwise, you’ll be limited to only a few revisions. Hence, why I find better results with my local LLM approach than on Xmind as a free user.

While there are many free ways to create mind maps online, if you want AI integration, I’d suggest using your own AI to generate the Markdown file with detailed prompts first, then uploading it to Xmind to turn it into a visual mind map.

Using locally generated Markdown with Xmind — Screenshot by Jayric Maning –no attributions required

Overall, I still prefer using my Obsidian and LM Studio setup mainly because it’s free and gives me more control over my custom prompts. However, I do understand why others might prefer paying for services like XMind instead of going through the trouble of setting up a local system and experimenting with prompts through trial and error.

Start simple, then level up

This local approach costs nothing, keeps your data private, and improves as you refine your prompts. Yes, the setup takes some effort upfront, but you gain complete control over your results. Your mind maps get clearer and more useful the more you experiment with different prompts and learn what works for your content. Start with the basic workflow I outlined, then tweak it to fit your needs. Try different models, refine your prompt templates, and adapt the system to match how you actually work. The potential here is massive because you basically control everything—it just takes time and patience.