Ollama and ChatGPT: local model setup on Mac

The tutorial explains how to run models locally through Ollama while using the ChatGPT desktop app on Mac. The speaker says this integration is Mac-only at the time of recording; Ollama also works on Windows and Linux, where other apps provide similar functionality.

The walkthrough starts with installing Ollama, enabling ChatGPT under Apps, and restarting ChatGPT. Model selection has a quirk: the speaker finds five entries already selected and must remove one before adding another in a text-box-like control. A second restart makes the local models appear beside OpenAI models in ChatGPT's picker.

The speaker says the interface and shortcuts remain familiar, but warns that switching an existing conversation to a cloud model can send its sensitive context along. He recommends starting a new chat when switching. His discussion of training, retention and app access presents his account of OpenAI's policies, rather than an independent privacy audit.

For an offline check, he suggests disconnecting the network and confirming that a downloaded model still responds. He also notes that models must fit in RAM and may load slowly on first use. His speed and quality comparisons are qualitative, with no benchmark results supplied. The setup applies per user; he has not evaluated the team and enterprise option he mentions.