March 15, 2026
This guide is coming next. It will show how to run Ollama on a Mac, use a strong local model like Qwen 3.5, and connect to it from Cumbersome on the same Mac or from an iPhone on your local network.
I am working on a thorough how-to for people who want local AI on Apple silicon without giving up a
decent client experience. The short version: Ollama makes it easy to run an open model locally on a
Mac, Qwen 3.5 is now good enough to be useful for real writing and coding work, and Cumbersome is a
much better front end than poking at localhost in a terminal all day.
The full post will cover:
- why you would run a local model at all: privacy, control, and keeping prompts on hardware you control
- what kind of Mac or Mac mini you need for this to feel good instead of miserable
- how to install Ollama on macOS
- how to pull and run Qwen 3.5 (and similar models)
- how to connect Cumbersome to Ollama on the same Mac
- how to use that same setup from an iPhone or another Mac on the same network
- how to clean up logs automatically if you want an even tighter privacy story
- a part 2 on using Tailscale for remote access without exposing the machine to the public Internet
One reason I want to write this now: local models crossed an important threshold. They are no longer just a toy for benchmarks and screenshots. A model like Qwen 3.5 can get surprisingly close to GPT-4o-class usefulness for a lot of everyday work, especially if your priority is privacy and "good enough" on a machine you already own.
If you want to follow along when the full guide lands, this is the URL I will build it on.
In the meantime, if you have not tried it yet, Cumbersome already supports OpenAI-compatible providers. That is the plumbing this walkthrough will use for a local Ollama endpoint.
Bless up! 🙏✨