Large language models (LLMs) aren't just chatbots or more advanced, occasionally hallucinating search engines, and there are many uses for them outside of talking to them. In fact, when treated like an actual language model, there are a ton of uses for them in productivity contexts that help me in my day-to-day life. From helping me be organized to making my day-to-day life easier, these are some of the ways I've deployed a local model to help me out.

5 Making Obsidian an even better note-taking app

My local LLM has my notes

I've been using Obsidian for quite a while now, and it's grown more and more to be the centerpiece of my digital life. Not only do all of my work-related notes live inside of it, but my projects, home lab adventures, and reverse engineering projects do, too.

With the help of an incredible plugin titled "LLM Workspace", you can integrate Obsidian with your self-hosted LLM. Define an embedding model and a chat model, then create a "workspace" with links to your other notes. From there, you can index all of those notes and query them from the Obsidian UI.

Note that for this plugin, the latest version in the community plugin repository does not have Ollama support. I had to download the plugin from GitHub and install it manually, but it worked perfectly once I set it up. I can query my notes and get ideas on how to expand on them further, or organize them in a better way. Plus, there aren't any data privacy concerns, as my notes aren't ever leaving my home network.

4 It powers my home through Home Assistant

A day summary and a weather report

A local LLM is the backbone of my entire Home Assistant setup these days, and powers a lot of my automations. From a summary of the work I need to complete for the day to custom weather notifications, it's invaluable at this point. Even my local voice assistant (which sounds like GLaDOS by the way, highly recommend), makes use of Gemma3 27B IT QAT for fast, high-quality responses.

This is one of the best benefits of an LLM, and it's so, so good. I can ask context-based questions about my smart home, and there's nothing "predefined" as such. Whether I ask if it will rain tomorrow, if I need sunscreen today, or if it will rain after 9 PM, I'll get an answer fueled by my weather data, and it's unique every time.

All of this not only saves me time, but it makes it so much easier to get a grasp of the day instantly. I get a summary of my calls, my workload, and if I need to be anywhere, complete with the weather for the day. Plus, keep in mind that I'm using a rather overpowered model for this use case. Even a small 1B model like Tinyllama or a 7B model will do the job; I just use Gemma3 for nearly everything else, so I may as well use it here, too.

3 Continue.dev is one of the best VS Code extensions out there

Great for a holistic overview

We've all heard of the "vibe coding" phenomenon, but the truth is that an LLM is still a great tool to use as a part of your programming workflow. So much so that tools like OpenAI's Codex, Gemini CLI, and Claude Code are making waves in the programming space thanks to their capabilities.

To be clear, I'd never rely on an LLM to do my programming for me, but it can be great to ask one if there are any issues with your code that it can immediately spot, or even use one to get a start on how to approach a problem. You should obviously do a quick sanity test and question whether the response makes sense in the context of your code, but it can be a great way to get you thinking about your code in other ways and the potential dangers it may pose. Continue.dev works with remote models or local ones, and I point it to my Ollama instance with Qwen-2.5:32B-Instruct.

For example, I developed my own software to manage the HUB75 64x32 Waveshare panel from an ESP32 and a Raspberry Pi. I can simply ask, within VS Code, if there are any issues with my code. It will then break it down and show me where there are potential problems while offering solutions for them. It's fantastic, and can be especially useful if I'm having an off-day when it comes to programming or I'm not as familiar with the language I'm writing code in. It may not catch everything, nor may its suggestions be 100% correct all of the time, but every bug or memory allocation issue that it does spot correctly means an instant improvement to my code.

2 Improving my home lab structure

It helps to know what you don't know

Sometimes, when I'm deploying a new piece of software or hardware, I need a plan to structure how to deploy it. I'm fairly adept when it comes to technology and self-hosting, but the nature of not knowing something is often that you don't know what you don't know.

As a result, it can be good to go to my self-hosted LLM, which has details of my home lab in a knowledge base in Open Web UI, and ask how a deployment would work in the context of the hardware and software that I already use. I don't take the answer as the 100% correct answer, but when you don't know what you don't know, a new term that is thrown at you or a self-hosting architecture you're unfamiliar with can spur a flurry of research into a new topic that you weren't aware of.

When it comes to LLMs, "trust but verify" is a great approach, even if "trust" is a bit of a strong word. It'll generally put you on the right track, but they'll often miss out on important details. Use it as a semantic search engine of sorts; describe what you're doing, and use the output to do research yourself on a topic you may be unfamiliar with. LLMs should be a tool to help you think rather than a replacement for thinking at all.

1 Email triage

Instantly get a summary of "important" emails

This is something I'm experimenting with currently, but using Home Assistant and Ollama, I'm attempting to build a triage of sorts that will flag whether an email should be considered "important" or not, using the IMAP integration in Home Assistant.

This experiment may ultimately fail, but the idea is rather simple. Extract the content of the email using Home Assistant, pass it to my local LLM, and get a response based on whether it meets my predefined criteria. If it does, it shows up on my dashboard (using a conditional card) so that I won't miss it. I could even summarize the collection of emails that have accumulated so far.

A local LLM is perfect for this; I don't want to share my emails with OpenAI, and a self-hosted solution gets around this potential privacy issue in a way that doesn't compromise on performance. A 27B model is more than enough for this task, and many smaller models will be capable of the same thing. In the long run, it'll save me time, but I'm still testing it at the moment.

Self-hosting an LLM makes all the difference

Sure, one could argue that ChatGPT and other cloud-based LLMs are great, but the truth is, local models are still incredibly powerful, and the power of those cloud-based options is, honestly, overkill in comparison. Self-hosting means I control the data, the context, and the integrations, and that's what matters most to me.