5 ways I'm using local AI to make the most of my old PC hardware

Published Oct 16, 2026, 7:00 AM EDT Nick Lewis is an editor at How-To Geek. He has been using computers for 20 years --- tinkering with everything from the UI to the Windows registry to device firmware. Before How-To Geek, he used Python and C++ as a freelance programmer. In college, Nick made extensive use of Fortran while pursuing a physics degree. Nick's love of tinkering with computers extends beyond work. He has been running video game servers from home for more than 10 years using Windows, Ubuntu, or Raspberry Pi OS. He also uses Proxmox to self-host a variety of services, including a Jellyfin Media Server, an Airsonic music server, a handful of game servers, NextCloud, and two Windows virtual machines. He enjoys DIY projects, especially if they involve technology. He regularly repairs and repurposes old computers and hardware for whatever new project is at hand. He has designed crossovers for homemade speakers all the way from the basic design to the PCB. Nick enjoys the outdoors. When he isn't working on a computer or DIY project, he is most likely to be found camping, backpacking, or canoeing. If you have an old computer (even a laptop) sitting around doing nothing, you could put it to use for local AI instead. I have two old PCs running various AI tasks all the time: one old HP laptop and a retired gaming PC with a Ryzen 1600X, 24GB of VRAM, and an RTX 3060. Selectively using local AI allows me to control what information gets sent to the cloud, create customized AI for my specific needs, and avoid running through my weekly Claude quota. The central piece of the setup is Ollama, which makes all of the AI models accessible on my local network. From there, other devices and services can reach the laptop or gaming PC and use them as miniature AI servers. The types of models you can run on this sort of reused hardware aren't as smart or capable as the huge frontier models, but a bit of careful planning can make them extremely useful on a daily basis. I made an AI Discord bot It can run on as little as 6GB of RAM My first big local AI project was a local Discord bot. It can play music, answer questions in voice chat, talk chat with you about any number of topics, look things up, act as a wiki, and more. It is significantly smarter than any off-the-shelf voice assistant, but it had one problem: it was very demanding. The full version of the Discord bot required two Blackwell GPUs to reach its full potential. Since then, I've created a more slimmed down version named Chad that will run on a modest gaming PC without an issue. It has three basic parts: Whisper — Handles speech to text. Llama3.2b — Parses casual spoken commands and maps them to a pre-defined list of commands. Komoro — Replies to the command out loud. Between those three components, Chad feels almost like having an Alexa sitting in Discord all the time without any of the privacy implications. For example, I could say "Queue up some Led Zeppelin" and it would know that means "Play Led Zeppelin," or I could say "I'm not feeling this song" and it'd know I mean "Skip." It can: Play music Manage a music queue Roll random numbers or pick a random person Grab voice clips on command Run polls Whisper isn't perfect. The larger AI Discord bot I run uses Parakeet instead, which offers moderate accuracy improvements. Llama3.2b also sometimes gets it wrong, but that is a small price to pay for a fully voice-controlled Discord bot. It also runs a larger Qwen model for a more intelligent conversational brain. Automatic photo tagging and search Immich makes for a great personal photo repo Credit: Ismar Hrnjicevic / How-To Geek Immich is a free and open-source replacement for Google Photos. If you have the right hardware, it supports face recognition and AI-powered search, which allows me to search for specific people or places using natural languages. A compatible GPU is critical for the AI features. Indexing a large library of photos is demanding enough, and trying to run a local AI on it in CPU-only mode is going to take forever. So, rather than run Immich on a high-power device that I don't want to leave running 24/7, I just run the AI container on the old gaming PC. While it is active, Immich can make use of the RTX 3060 for AI jobs; when it goes offline, Immich still works as a regular photo library and can lean on the work the RTX 3060 already did. Still, if you import 30,000 photos (about how many I have on my phone), you should expect it to take a while, even on a beefy PC. Sorting and naming scanned documents Paperless-ngx makes the mountain of paperwork disappear I keep most of my important paperwork even if it isn't "necessary." Unfortunately, that also means I have a huge backlog going back years. Over the last year or so, I've been slowly adding it to a Paperless-ngx server, which serves as a digital repository that allows me to search for documents based on the contents in the document. Recently, Paperless-ngx integrated local AI features too. Now, it is capable of using AI to title, tag, and date the documents you upload. Because the entire thing runs locally, I don't have to worry about a tax document or my passport application leaking to the web. Everything is safely secured on my own server. . A local voice assistant for my smart home An AI Discord bot and a voice assistant are very similar Credit: Justin Duino/How-To Geek If you're running any kind of AI assistant, it is usually pretty easy to turn it into a replacement for Alexa. In my case, I just adapted my lightweight Discord bot, Chad, to serve as a Home Assistant too. All of the AI components remained exactly the same. It uses Whisper for speech to text, llama to parse natural language to commands, and then komoro to reply to me out loud. I just needed to update the bot, which needs to interface with Home Assistant to control different parts of my smart home setup. Currently, the voice assistant can control the lights, set alarms and timers for me, and read back notifications from a few apps. It runs on an old laptop, so there is a bit of a delay when it speaks back to me. If you could run it on any RTX GPU, it'd be very close to real time. The next upgrade will integrate it with Mealie so it can answer questions about recipes as I cook, and allow it to use satellites, so I can use it in any room in my home. A screenshot helper for any device on the network AI labeling without the fuss I frequently work between my desktop, my laptop, and my phone. On my desktop, I run a small local AI that automatically labels screenshots for me as I take them. My phone and my laptop, however, don't have that capability. To make working on my phone or laptop less tedious, I set up a small AI model that automatically analyzes and renames screenshots as I take them. That way, when I need it later, I don't have to scroll through 1,000 screenshots—I can just search for the app and what I was doing to find it instead. It occasionally makes mistakes, but it has become an invaluable part of my workflow that I can now use on any device on my network. An old PC is great for AI An old PC—even a laptop—can be quite useful for AI. The trick is to keep the AI jobs focused. Don't use a 27 billion parameter model on a laptop 16GB of RAM and no discrete GPU, it is just inviting trouble. However, a few 3-5B parameter models that are optimized to run on a CPU are perfectly reasonable. I have a few that run in near real time and a few other, larger models that are basically designed to be left running overnight. If you have an old desktop with a dedicated GPU, you have even more options. They can often run a handful of small AI models faster than real time or one medium model in real time. If you're willing to accept CPU offload and slower processing times, you can even run models like Qwen 3.8-27B, which is a fantastic coding and reasoning model for its size.