A developer built an AI code editor that lets you choose the model, after Cursor's Grok default

What the author says changed in Cursor

As dev.to reported in a post by a developer, Cursor, a popular AI code editor, has been steering users toward Grok since SpaceX bought the company in August. Grok is the AI model from Elon Musk's xAI. Anyone who opens the editor may notice their default model has switched.

The post adds a pricing claim. Plans starting 12 October, it says, include usage for Grok only. Other models, Claude among them, would be charged at the provider's published API list price from the first request. The API is the paid channel software uses to call a model directly.

These are the author's claims. He points to a forum thread and an article as his sources, and we have not checked either one. He also says he is not arguing about which model is best. His point is that the user should decide.

How Quill lets you pick the model

The editor, called Quill, offers four routes to a model. The first is local models: it detects programs such as Ollama, vLLM and LM Studio, which run AI models on your own computer, and it can work with no internet connection.

The second is a Claude command-line tool you already have installed. Quill notices it and works through your own Claude account, so the author adds no charge on top.

The third is your own API keys, which are password-like codes that let software use a provider's service. Keys for Anthropic, OpenAI and Gemini work, they are kept on your computer, and they do not count against your Quill plan.

The fourth is hosted models run through Quill's plan: Claude, GPT, Gemini and Grok. There is also an Auto setting that chooses for you. You can set a different model for autocomplete, for chat and for the agent, which is the mode that carries out multi-step tasks on its own.

What else the editor does

Quill has five working modes: Ask, Plan, Agent, Multi-Agent and QA. Multi-Agent divides a large job among helper agents, each working in a separate copy of your project. QA, available on the Pro and Max plans, takes a test written in everyday language, carries it out in a real web browser, and reports pass or fail for each step.

The author's answer to AI forgetting what your app is called the Feature Tree. It is a map of your app's modules, pages, actions and fields, kept in a file named feature-tree.yaml. The agent consults it in each chat. If the agent wants to change the map, it shows the change as a diff, a view of what was added and removed, for you to approve.

That approval step applies everywhere. Each AI edit shows up as a diff, and your files are not altered until you accept it. Quill also imports your VS Code extensions, themes, settings and keybindings, which lowers the cost of switching.

On money, each reply shows what share of your plan it used. You can set warnings and a daily spending cap.

Price, privacy and the fine print

Pro costs $20 a month and Max costs $40. New users get a free trial of all Pro features, and no card is needed. On privacy, the author says he records usage only, and never your code or your messages.

The author is open about his bias: he built the product he is promoting. Only the pricing and the Cursor claims are in his post, and we have not tested Quill or confirmed his account of Cursor's plans. Treat it as a maker's pitch, not an independent review.

He also asks readers what is missing and directs requests to a forum on his site, which suggests a young product shaped by early users.

The lock-in question underneath

The larger issue is what happens when the company behind your tools picks a favorite model. If the default changes, or if other models become more expensive to use, your habits and budget change with it. You may not have asked for either change.

A tool that supports many models spreads that risk. You can compare the same task across models and keep the one that works best. You can also switch when a provider's price rises or its service slows. This ties into today's other stories about tight computing capacity, where relying on one provider can leave you waiting.

The tradeoff is real. A newer editor from one developer may lack the polish, support or longevity of an established one. Picking your own model also means managing keys and costs yourself.