Forget the complicated guides: running local LLMs is easier than you think
Lessons from installing local LLMs on an old Mac — what works, what doesn’t, and why simpler is better.
Running a large language model (LLM) locally on an old Intel MacBook? At first, it sounded unrealistic.
My goal was simple: have a private, offline chatbot — something like ChatGPT — running entirely on my own machine, with no cloud dependency, no data sharing, and ideally with a smooth user interface.
The motivation? Privacy, experimentation, curiosity — and the fun of pushing old hardware to do modern AI tasks.
My test machine: MacBook Pro Retina 15”, Mid-2015, Intel i7, 16 GB RAM, macOS Monterey. A 10-year-old laptop that still works like a charm for writing and light dev — but by no means a powerhouse.
Could this Mac really run an LLM? And how complicated would it be?
The web was full of tutorials — but many seemed outdated, convoluted, or conflicting. So I did what many users would do: I opened ChatGPT and asked for help.
What followed was an experience of trial and error, multiple installs, conflicting advice — and eventually, a very simple solution that I wish I had known from the start.
This article walks through that journey: the process, the pitfalls, and a step-by-step guide for anyone looking to run an LLM locally on a Mac. Especially an older Mac.
First steps — and first confusion
The first thing I did was head to Google:
“How to run an LLM locally on a Mac?”
As expected, I was bombarded with guides — some old, some incomplete, many written for Linux or Apple Silicon Macs, not Intel. Some suggested compiling llama.cpp, others insisted I needed Docker, Python, Homebrew, Node.js. A few mentioned Ollama. Fewer still explained how it all fit together.
So I turned to ChatGPT for guidance.
The initial advice was… elaborate:
I was told to install Homebrew, Python 3.13, Node.js, llama.cpp (compile from source), and the AnythingLLM GitHub version — to run a Node.js server manually. Then download GGUF models, write a Modelfile, create models inside Ollama, and run both the server and Ollama in tandem.
It sounded doable — but unnecessarily complex for what I wanted: just a ChatGPT-like interface running locally.
Even worse, some of the steps didn’t apply well to macOS Monterey or Intel Macs. The advice leaned heavily toward setups that work on Linux or Apple Silicon, not older Intel hardware.
Pretty soon I had multiple tools partially installed, folders scattered, and growing doubt about which path would actually work.
Multiple installs & troubleshooting
I went down the rabbit hole.
Updated Homebrew — useful on a Mac, no harm.
Installed Node.js — multiple times, due to version issues.
Tried upgrading Python — I had multiple installations (with and without Homebrew), so this required quite a bit of cleanup, and then I realized I didn’t need it. But at least I now have a clean setup for Python!
Cloned and built llama.cpp — which worked, but wasn’t what I wanted for daily use.
Downloaded the AnythingLLM GitHub version — and this is where the real problems began.
I moved the GitHub folder as advised, only to discover that I was missing key files. Running npm run dev failed. I kept getting “missing package.json” errors.
ChatGPT kept pushing me toward the GitHub version — but what it didn’t tell me: AnythingLLM now has two different versions:
- A GitHub project (requires manual setup)
- A Mac app version — installs like any app and manages everything for you
That distinction would have saved me days. Once I realized it, the solution was obvious.
What actually worked — the simple path
After all the detours, the solution was simple:
✅ Ollama.app
✅ AnythingLLM.app
🚫 No Python. No Node.js. No terminal commands. No coding.
I launched Ollama.app, which runs in the background and handles model inference. I opened AnythingLLM.app, which detected Ollama and let me select and download models.
That’s it.
What I learned (so you don’t have to repeat it)
Here’s the key takeaway:
Running an LLM locally is much easier than most guides suggest — if you pick the right tools.
Most web content — and even ChatGPT — still lean toward “build it yourself.” That works, but it’s overkill for 90% of users.
For a basic, ChatGPT-like offline experience:
👉 Ollama.app + AnythingLLM.app = enough.
You don’t need to touch Python or Node.js — and you don’t need to be a developer to set this up.
Most of the guides online are outdated. In 2024, GitHub builds were common. By mid-2025, packaged apps are more mature and user-friendly. Many tutorials haven’t caught up, and all the old ones are still out there.
And yes — Intel Macs can still run LLMs. My 2015 model handles Mistral-7B uncensored Q4_K_M with no problem. Just choose your model wisely.
Step-by-step guide: run Ollama + AnythingLLM app on your Mac
My Setup:
- Mac running macOS Monterey or newer
- Intel Mac (works fine — mine is a 2015 MacBook Pro 15”)
- 8 GB RAM minimum — 16 GB better
- 5–15 GB disk space per model
What you will install:
- Ollama.app — runs models
- AnythingLLM.app — provides ChatGPT-style UI
How to install:
Step 1: Install Ollama
- Go to https://ollama.com
- Download and install Ollama.app
- Launch it (it runs in background)
Step 2: Install AnythingLLM
- Go to https://anythingllm.com
- Download AnythingLLM.app
- Install and launch
Step 3: Download a model
- In AnythingLLM, select Ollama as provider
- Download a model (Mistral-7B Q4_K_M is a good starting point)
- Set model name, and chat away — fully offline
Appendix
App version vs GitHub project:
RAM & model size tips:
Recommended models for Intel Macs:
- Mistral-7B Q4_K_M → best balanced
- Nous Hermes 2 Mistral 7B Q4_K_M → uncensored, creative
- LLaMA 3.1 8B Lexi Uncensored → heavy but possible
- Phi-2 2.7B → lightweight, very fast
Tips for better performance:
- Only run 1 model at a time
- Close heavy apps
- Stick to Q4 quantized models
- Keep Ollama.app and AnythingLLM.app updated
If you found this useful, feel free to share or comment — and let me know what models you’re running locally!
