VibeCode Studio

I created a tool for coding with LLMs, mainly for Apple platforms development (but I guess you can use it for anything else). It is free on appstore for Mac and it is called... VibeCode Studio. What it does is, it allows you to select only relevant parts of your codebase (assuming you are organized enough to have it structured in different files) and assembles those into a prompt which gets copied into a clipboard and that you can use with any LLM. Plus it keeps tracking of your selected files, their length, and it allows you to adjust your prompts. Few tips: - Instead of using integrated tools, I prefer to keep it separate and copy/paste parts of my code to different LLM chats - I tried many LLMs, but Claude wins - Unless a specific, complex problem that cannot be solved otherwise, usually I get better results without reasoning. And it is way cheaper and way faster - Claude 3.7 is available on Perplexity. I got a deal from my mobile provider (T-mobile) for a year of free Perplexity subscription. Their UI is also pretty OK, so it is my default go-to platform. Just use Claude 3.7 without reasoning and without websearch - Before I used Anthropic Console, which also works well. I always resort to it when Perplexity has some issues (happens sometimes) - Instruct the LLM to give you whole structs/classes/functions so that you do not have to copy/paste around placeholders like // ...rest of your code goes here - Always version control, commit often with good comments! You need to have a reliable way back - Code folding! In Xcode this works beautifully for replacing whole code blocks, in other editors like JupyterLab it also works nicely, so turn it on and use it Unlike the common anecdote, for me vibecoding is actually enforcing good practice relatively soon in any project. You start with one file and once it gets clumsy/expensive, it forces you to split into different files. It also forces you to split functions/classes as having functional modules helps a lot with your prompts and with integrating the LLM output. Also, as you move fast and try different things, it forces you to use git in quite an extensive way. Happy coding!