I stopped using ChatGPT, Claude, and Gemini for every little task, and my local LLM picked up the slack

Published Oct 30, 2026, 12:31 PM EDT Mahnoor Faisal is a tech journalist covering AI and productivity tools with bylines at XDA, SlashGear, MakeUseOf, Laptop Mag, and Android Police. She's been writing professionally since she was sixteen, and has since penned hundreds of articles. This includes in-depth coverage of AI tools like NotebookLM to breaking news across the AI space. Her passion for technology started when she received her first iPod Touch (4th generation) on her 8th birthday, and she's been deep in the tech world ever since. Currently pursuing a degree in computer science, Mahnoor brings both a journalist's eye and a technical foundation to her coverage of how AI is reshaping the way we work and learn. Unlike subscriptions like Netflix, Spotify, or Apple Music, where paying a monthly fee generally means you can use the service as much as you want, AI subscriptions don't quite work that way. You could be paying $20, or even $100, a month and still find yourself staring at a polite little message telling you that you've hit your usage limit and should come back later. The irony is that AI is probably one of the most expensive subscriptions you pay for, yet it's also one of the few that can lock you out despite paying for it! This is one of the reasons (among others) I've stopped turning to ChatGPT, Claude, and Gemini for every little task. Instead of wasting my limited usage on tasks that barely scratch the surface of what these tools can do, I started handing them off to a local LLM running directly on my devices. We've gotten a little too comfortable using AI for everything Not everything needs a powerful AI model Remember when Googling something was practically a reflex? Whether you had a random question, needed to understand an unfamiliar term, or wanted to settle a debate with a friend, your first instinct was probably to open Google and search for it. Well, I think AI chatbots have slowly started taking Google's place in that regard. It's unfortunate, but it's true. Instead of typing a question into a search engine and scrolling through even a single result to find an answer, it's often far easier to ask ChatGPT, Claude, Gemini, or whatever your preferred chatbot is and get a response directly. It doesn't just stop at questions! We've gotten so comfortable with having AI at our fingertips that turning to a chatbot for even the smallest tasks has become almost second nature. I've seen people turn to AI to decide if their outfit looks good or not! The problem is that every one of those requests counts toward your usage limits, regardless of whether you're asking AI to solve a complicated problem or simply rephrase a sentence. Beyond the frustration of running into those limits, there's another reason I've started questioning how frequently I rely on cloud-based AI: its environmental impact. Running powerful AI models requires enormous amounts of computing power, and the data centers behind them consume electricity and, in many cases, substantial amounts of water for cooling. As demand for AI continues to grow, so does the infrastructure needed to support it. While I'm certainly not suggesting we should stop using cloud AI altogether, I do think it's worth questioning whether every little task really needs to be sent to a massive data center halfway across the world. Think about what you're actually using AI for It's the little things that add up I know I'm writing this as a journalist, so what I'm about to type after this hurts my soul and doesn't really apply to me on a personal level. But if you aren't a writer, there's a good chance you've become somewhat reliant on using AI for writing help. I'm not saying you do, or that you shouldn't, but I say this based on the countless conversations I've had with people. Even AI companies frequently use writing-related tasks as examples when demonstrating what their models are capable of. Now, there are two types of writing tasks you'll see people turn to AI for. One is the full-fledged "write me an SEO article about XYZ" kind, which requires research from the web, fact-checking, and the ability to put together a coherent, well-structured piece. That's the sort of task where having access to a powerful cloud-based model can make a difference. The other is much simpler. Think rephrasing an awkward sentence, fixing grammatical errors, making an email sound more professional, or shortening a paragraph that's gotten a little too wordy. These tasks don't require extensive research, complex reasoning, or access to the latest information. Do you really need one of the most powerful AI models in the world to handle any of that? I've been using my local LLM as a Grammarly replacement for months! Cleaning up existing text is, of course, far from the only example. I'm currently a student, so a lot of my AI usage naturally gravitates toward using it as a tutor. Beyond relying extensively on tools like NotebookLM to study, I also almost always have a chatbot open in another tab so I can ask questions whenever I get stuck on something. Sometimes, that means asking it to walk me through a particularly confusing concept. Other times, it's something as simple as asking what a term means, requesting another example, or getting a quick explanation of something I didn't quite understand the first time around. While there are times I need the advanced reasoning capabilities of a powerful cloud-based model, not every question I ask requires that level of intelligence. For straightforward definitions, explanations, and examples, a smaller local LLM can often do the job just fine. Another example is the relatively simple tasks you might've automated for yourself using AI. For instance, you might have an automation that summarizes your most important emails or gives you a breakdown of what you have going on for the day. Or, you might've built a custom tool specifically for your workflow. I personally have a skill that automatically renames any screenshot I take, giving it a descriptive filename based on what's in the image rather than leaving it with the generic name macOS assigns by default. While I initially had this working through Claude by building a skill I'd run every time I needed to rename a screenshot, I've since swapped it for a script-based implementation that uses Gemma 4 running locally on my Mac. The end result is essentially the same, except I no longer need to rely on a cloud-based model or use up my Claude usage limits every time I want to rename a screenshot. I run local LLMs on both my phone and MacBook Local AI doesn't have to be complicated While I'd love to use my devices to run local LLMs for more demanding tasks, I'm currently working with a MacBook that has just 8GB of RAM. This means outsourcing the little tasks is pretty much all I can realistically do. I've tried a fair few local LLMs that work well with my devices, but Gemma 4 is what I've ultimately settled on. The smaller variants are lightweight enough to run on my hardware without bringing everything else to a crawl, while still being capable enough to handle most of the everyday tasks I've discussed above. I particularly enjoy using a local LLM on my phone. The fact that I can open Google AI Edge Gallery just as easily as I'd open ChatGPT, Claude, or Gemini, and ask it whatever question I have without needing an internet connection or worrying about usage limits, is incredibly convenient