Vibe Coding an API with Magic

For 3 months we've worked on our own LLM. It's based upon OpenAI's gpt-4.1-mini, but it's a completely proprietary thing, and it allows us to generate backend API code from natural language.

We're not completely done yet, but being an open source company, we're of course building in public.

What is Vibe Coding?

Vibe coding is a new word, and it implies that instead of manually creating all required code, you use AI to generate a large part of it. We've done this for years already, using tools such as ChatGPT, Microsoft CoPilot, and now recently Cursor AI and Manus AI.

The idea is that if you can phrase your questions using natural English, the AI can for the most parts transform your English into a working API. You can actually try it for yourself here without any downloading required, by visiting the following link, click the chatbot button, and ask it the following.

Generate an HTTP endpoint that allows me to send an email using [email], [subject] and [body] arguments

The above question should generate something resembling the following ...

Vibe Coding in Magic

We've been going "all in" on vibe coding at AINIRO. The problem we're facing though, is that Hyperlambda is simply too narrow for existing LLMs to be able to understand it. So we were forced into creating our own LLM that understands the syntax.

For the last 3 months we've created approximately 16,000 Hyperlambda files. It's a dirty job, but since LLMs basically learns from examples, there's no way around it. At the time being it's got a "medium good" understanding of the language, but we create new training material every single day for it, to strengthen its understanding. So in a couple of months we expect it to be at least close to "SOTA" (State Of The Art) when it comes to backend development, APIs, and stuff that "runs on your server". For the record, we have no plans to implement support for generating frontend code. Magic is, always has been, and always will be about backend software development. If you need frontend generators, you'll need to go somewhere else.

Try it out for yourself

Magic Cloud is open source, and you can fork it here. And at least for now, we'll be paying for your tokens too as you generate Hyperlambda. We plan on changing this in the future, but for now we're giving it away for free. If you want to download and try it, you should use Hyper IDE or the Hyperlambda Playground to generate code, such as I illustrate in the above video. Screenshot below ...

Where the point being that you write the comment, and Magic creates the code ...

In addition you can try it using our documentation chatbot, just remember to start out your questions with "Generate Hyperlambda", at which point it will trigger our Hyperlambda Generator LLM to create your code.

It is still not 100% ready, but since we're building in public, we want you to try it out and give us feedback.