Local LLMs aren't just for messing about and having fun with; they can genuinely be useful tools, too. While there are certainly more obvious use cases if you have a home lab, LM Studio enables you to do a lot of the same stuff without needing to know how to set up Ollama or another alternative LLM hosting solution. With its OpenAI-compatible API, many services that support OpenAI but allow you to designate a custom endpoint can be used this way, and there are many ways to make it work for you.

I've been using local LLMs for quite a while now, and I have both an Ollama instance running on my server alongside an LM Studio server running on my main PC. LM Studio is just as good, and for many of my day-to-day workflows, it's become a core component of how I use my computer. Here are some of the ways that I use it on my PC.

Alt text generation

Vision models can help you make the web more accessible

Alt text is a very important accessibility feature on the web, and it adds a special tag that describes the contents of an image. Not only does it allow crawlers and web scrapers to understand what each image depicts, but it also allows those who use screen readers to understand the images that are on the page. Thanks to LM Studio and Gemma 3 27B, I can generate alt text for any image on a page and have it copied to my clipboard immediately.

Of course, it's not always perfect. The above image is accurate for the most part, but even when it comes to identifying an ESP32, it says "Black circuit board with gold pins, chip, and connectors." It's not wrong as such, but it's definitely not as accurate as it could be. It's a good start, though, and you can tweak the text to better fit your image.

If you publish images online at all, be it on your own website or on social media, alt text is everywhere. It's good to use it where you can, and LM Studio and a local LLM capable of vision make it a whole lot easier. Plus, because it's local, your images don't ever leave your computer until you upload them.

Make your note-taker even better

Many tools have plugins for language models

Obsidian has become central to managing nearly all aspects of my digital world, from work documentation to personal projects like home lab experiments and reverse engineering endeavors. At this stage, I'd probably be lost without it, and I have organized notes going back over the last year in there that help me keep on task, plan for the weeks ahead, and more.

The "LLM Workspace" plugin takes this a step further by enabling integration with an LLM, including support for locally-hosted options. After specifying your embedding and chat models, you create a workspace linked to existing notes. This allows you to index everything and query it directly within Obsidian's interface. There's also the official Obsidian "Copilot" plugin, though there's no affiliation with Microsoft.

A quick note: the newest version of the LLM Workspace plugin available through Obsidian's community plugins lacks Ollama support. I needed to install directly from GitHub, but once configured, it functions flawlessly. With it, I can easily brainstorm expansions for my notes or refine their organization, and all while keeping my data private on my own network.

A coding buddy that's better than Copilot

Continue.dev is perfect for this

I'm not the biggest fan of Microsoft's Copilot integration in Visual Studio Code, and there are a couple of reasons for that. For starters, the free version is limited to 2,000 auto completions a month and 50 manual requests. That's not a lot, and if you program a lot, you'll burn through the auto-completions quite quickly. On top of that, while all data is excluded from training by default, I don't need all of my code being sent to servers. It's not that I care as such, but it feels completely needless.

Instead, I can use a local language model with Visual Studio Code, either through Continue or using the model picker that's now built into the IDE. This has complete autocomplete support, the data never leaves my PC, and I can initiate a conversation with the model from within Visual Studio Code, too. To get it working with LM Studio, all I had to do was add the following to my config.yaml:

- name: Qwen 3 30B
 provider: lmstudio
 model: Qwen/Qwen3-Coder-30B-A3B-Instruct
 capabilities:
 - tool_use
 roles:
 - chat
 - edit
 - apply

Document summarization

If it fits in the context window, your LLM can see the entire thing

This one is really simple, but I absolutely love it. You can drag a PDF, DocX, or TXT file into an LM Studio chat, and the app will use that file as context for your questions. If the file fits the model's context window, LM Studio inlines the whole thing; if it doesn't, it automatically switches to retrieval-augmented generation (RAG) and pulls only the most relevant chunks.

You can ask for a summarization, ask for specific information, or even use it as context for something related but different. With a large enough context size, you can basically have a conversation with your document, and even without a big one, you can still have LM Studio surface the relevant information from the documents if you need it.

While it's not suited to searching through a trove of documents, it can be great to find whether or not the information you require is in a large document. With RAG, it will search for the information you requested and show you it if it's there, all without filling the full context window.

An offline study helper for textbooks and research papers

Like NotebookLM, but local

If you've used NotebookLM before, this is kind of like that. NotebookLM is a Google tool that analyzes your uploaded sources to help with summaries and help you formulate questions and answers to test your knowledge. With LM Studio, you can recreate that experience offline by attaching chapters or even entire papers and prompting for structured outputs (like outlines, flashcards, and quizzes) without sending content to the cloud.

This can also work well with a note-taker like Obsidian. As we already mentioned, you can use an embedding model and an LLM with Obsidian to "talk" to your notes, so if you have a lot of notes, then you can use that integration to access your entire corpus of data using RAG.

Because LM Studio keeps everything on your machine, this is ideal for coursework and unpublished drafts, and will work even without internet. It's obviously not going to be as powerful as what Google offers, but it gets the job done and can be great for even coming up with basic questions for your studied content.

Add LLM-based automation to your home

Especially good for private data

LM Studio has two separate REST APIs that you can avail of. The first is an OpenAI-compatible API that mimics what you'll find on OpenAI, and the second is an LM Studio-specific API that offers additional telemetry and model info. Most people will only need the OpenAI-compatible API, and you can use it with most services.

To give you an idea of what's possible, Ollama also offers an OpenAI-compatible API, and I've used that with Home Assistant to generate email summaries. This is particularly beneficial, as my emails never go to an external service aside from the service they're initially hosted on. For privacy reasons, a local LLM is perfect for this, and you can do a lot more, too. For example, you could pair it with Frigate to identify what's happening in your camera feed at any moment.

Even the alt text generator I mentioned earlier is made possible through the LM Studio API that I'm using for my extension. It's incredibly powerful, and it's all on-device, on my home network, and works even when I'm totally offline.