Original video in Chinese.
Key Takeaway
- Ollama is now natively embedded in Codex: the latest version of Ollama supports one-click integration, and both official models and Ollama’s cloud/local open source models can appear in Codex’s model list at the same time. In the same conversation, you can switch models at any time, keep the context continuous, and swap out the underlying model.
- Configuration only takes two steps: completely close Codex, including its background processes, then turn on Ollama’s Codex switch, add the locally installed model in settings, and restart. The barrier is extremely low, and even beginners can do it.
- Open source models are a key piece of sovereign AI: open source models like Qwen and DeepSeek are already usable for everyday production, and the hardware requirements are affordable for individuals. Using both closed and open source systems is, in essence, about increasing choice rather than being locked in by a single provider.
You definitely need to install the latest version of Ollama. Because it now supports Codex embedding. In other words, you can very easily use open source models on Codex.
Look, in Codex’s model list, the ones below are models officially provided by OpenAI, and the ones above are models from Ollama, including their cloud models as well as the models I’ve installed locally myself.
And now, when using third-party models, all of these historical chats on the left are still there; it no longer becomes a new record like before.
While testing, I also discovered one more thing: you can switch models freely within the same conversation.
Look, I started with a local model. Then I switched from qwen3.8 to GPT-5.5, and then to gemma4. During this switching process, the conversation went on normally. Finally, I asked it what I had said earlier. It still had the previous chat history.
This means that the context data is shared, while the underlying model can be swapped at will. For those of us who want to build a personal sovereign AI system, this is very, very good news!
To get Ollama embedded into Codex, you only need to do two steps. Very simple. Even a beginner can do it.
First, completely close Codex. Then turn on Ollama’s Codex button.
This step may produce an error. I even had Codex help me check the cause. It turned out that Codex had not been completely closed. So you need to open Activity Monitor and shut down all Codex and ChatGPT processes in the background. Only then can Ollama finish the setup.
Second, go to settings, scroll all the way to the bottom, and configure the model you want. Then restart Codex. After that, the model list will include the open source model you want.
In the initial state, there are only Ollama cloud models in there, and the five slots are already full. What you need to do is delete the ones you don’t want, then find the locally installed model in the pop-up window and add it in—that’s it.
Seeing this, I know someone will definitely ask: is it necessary to mess with open source models?
First, open source models are already very capable.
The latest Qwen and DeepSeek prove that open source models can already be used for everyday production, and the hardware requirements are also affordable for individuals. So now I can call both closed and open source systems at the same time, bringing out the strengths of both and putting them to use for me.
Second, Ollama further lowers the barrier to entry.
Before, integrating open source models into Codex or Claude Code was pretty troublesome. Now Ollama has removed those barriers directly. Even beginners can set it up very easily.
Third, isn’t it good to have one more option?
I subscribe to the $100 Codex and Grok Bot, but I also use open source models. I’m not just trying to use AI; I’m trying to build a personal sovereign AI system.
Yesterday, I also said in Knowledge Planet that over the past three years, all the things I’ve done were, at their core, about wanting more choice—whether it’s using AI or living life well, that has always been my basic pursuit.
OK, that’s all for this episode. If you want to understand AI, if you want to reclaim individual sovereignty, if you want to find like-minded people, come to the newtype community. See you next time!