Running a local AI chat without internet access

May 20, 2025 · 1 min read

Originally posted on LinkedIn, May 20, 2025.

I was sitting on a train with no cellular reception and using an AI chat. Somehow, in 2025, there is still no continuous cellular coverage between Beit Yehoshua and Tel Aviv.

My workaround is a ChatGPT-like setup that runs without an internet connection. It takes two simple steps:

Install the Ollama app for Mac. Installation takes one click, followed by downloading a free model package such as llama 3.2.
Install Open Web UI with Docker. It takes one command in the terminal.

That gives me a ChatGPT-like interface in the browser, running entirely on my machine.

There are two limitations:

The model I tried did not support Hebrew. Another model might, but I have not investigated.
The answers are lower quality than those from the well-known models.

But it is private, fast, and available anywhere, including on a train stuck without reception.

Illustration for “Running a local AI chat without internet access”