I’ve been using Obsidian for several years now, and it’s been one of the most solid and consistently reliable options for storing my notes and building a knowledge base. This is where I keep long blocks of research, my novel drafts, quick notes, everything. Obsidian’s plugins can significantly improve the user experience, especially Copilot - it basically turns your vault into an AI-powered second brain.

As I’ve been switching more of my productivity stack to a local setup, and since Obsidian is very local-forward, I figured there must be a way to connect it to my local LLM instead of using the cloud models in Copilot. And it turned out to be dead simple to do. Now I’m not only reaping the benefits of an AI layer sitting on top of my vault, but it’s completely private and under my control too.

Why even set up Obsidian with a local LLM?

The pros outweigh the cons

The Copilot plugin, by default, provides you with a sizable list of AI models to choose from, including models by OpenAI, Google, and Anthropic. The snag is that you need to provide your own API key to use any of them. Another issue is that many of them aren’t free, and the ones that are free have limited credits. So out of the gate, there’s already the issue of cloud dependence and cost. These models get access to your local notes, and even if you’re fine with that, you’ll need to pay if you want to get any real use out of them.

A local LLM solves this completely. Hooking up Obsidian to a local LLM means none of your personal notes are accessed by anything that lives in the cloud, so you retain full privacy. The entire setup lives on your device, so it’s also usable offline. Although there will be some setup required on your part, there are no API keys or costs involved, and the configuration is extremely quick and simple; anyone can do it.

How I set up Obsidian’s Copilot with my local LLM

Getting Obsidian to talk to my local AI

The first thing I did was get a local LLM up and running. I already did this a while ago when I set up LM Studio. It’s one of the easiest runners and doesn’t require any coding knowledge thanks to its simple graphical interface, but some other options include Jan AI, Ollama, and AnythingLLM. All you do is install it on your machine, then look for and download your model of choice. Also ensure that you meet the hardware requirements; it will depend on the size of the model you’re running.

In LM Studio, I opened my model of choice, OpenAI’s open-source gpt-oss-20b. Then I went to the Dev tab (represented with a green >_ icon), and turned on the “start running” toggle. In Obsidian, I made sure the Copilot plugin was installed and enabled. Then I went to the Copilot settings, opened the Model tab, and clicked Add Model.

For the model name field, make sure you add the model name exactly as it reads in LM Studio or your runner of choice, character-for-character. Then select your local runner from the Provider list, and for the display name, you can call it anything. In the Base URL field, paste this address, http://localhost:1234/v1 (only works if you’re using LM Studio), and since local LLM runners don’t require API keys, just type “local” in the API Key field. Make sure you’ve got CORS ticked on, and then you can verify and add the model. And that’s it, Copilot was hooked up to my local LLM through LM Studio.

How I’m using my local LLM in Obsidian

It works the same way as any other cloud LLM

Using my local LLM in Obsidian is no different than any other cloud model you’d run through Copilot. For starters, using a local LLM through Copilot still fully supports retrieval augmented generation (RAG), so you can easily get it to retrieve, extract, and summarize key points from your notes. I also like the quick-access Copilot option in Obsidian. All I have to do is highlight text, right-click > Copilot, and select one of the template prompts such as “explain like I’m 5” or “simplify”. And whenever I want to have a back-and-forth with my notes, I open the Copilot panel, add the relevant notes as reference, and prompt it for insights, patterns, quizzes, and more.

Copilot does override a lot of the model’s behavior that you’ve configured in LM Studio, so I recommend setting the temperature, response length, system prompt, and so on in Copilot instead of LM Studio. The answers will also vary depending on your local model of choice, since each one has different training data and capabilities. That’s why it’s important to pick one that’s best for your area of expertise or domain. For example, my gpt-oss 20b is great for general-purpose querying, while Qwen3-coder is better for coding assistance, and Mistral 7b is good for creative writing.

AI in my notes, without compromise

Using an AI plugin will change how you interact with your notes and boost productivity regardless of whether it’s local or not. But hooking up Copilot to a local LLM comes with many benefits. The cloud doesn’t get access to your notes, you can use the entire system offline, and it’s completely free. There’s no downside to it except maybe spending ten minutes to set everything up.