Back to blog

Wednesday, November 26, 2025

We Just Shipped Real-Time Voice to RA-H

cover

Why This Matters

I find myself increasingly taking articles, podcast transcripts, and ideas and putting them into real-time voice streaming parts of language models. Grok voice mode, ChatGPT's voice features - there's something happening here beyond just "nice to go for a walk."

Your brain has a word queue. When you think internally, you can jump around. You don't have to piece together robust ideas. When you're writing, you can procrastinate, edit, leave blanks.

But when you're talking in real-time, you have this buffer where you're forced to piece words together logically and coherently. It reveals gaps in your thinking really quickly.

The Phonological Loop

This is called the phonological loop - part of your working memory that handles speech and sound. It's like a little voice in your head that repeats sounds or words, but it enforces a serial buffer. Speaking commits words in order, forcing you to bridge gaps. Writing lets you jump around.

That's why voice makes thinking more coherent.

Why We Brought It Back

I originally had real-time voice in RA-H a year ago, but the open-source voice models were terrible. Now they're really good. Nvidia has Parakeet, there are amazing voice-to-text and text-to-speech services.

I was using SuperWhisper (great app), but it doesn't force this loop - this forcing function that happens when you speak in real-time. You can stop, edit, jump around.

The Demo

In the video above, I show how I actually created the original voice interface blog post. I didn't write it - I sat in the lounge room with coffee and spoke through it. The system pulled in related articles and ideas from my knowledge base and fleshed out the piece bit by bit.

Same benefits you get speaking to Grok or ChatGPT, but connected to your personal knowledge base.

Try It Free

There's a free version that gives you access to all of this. I don't have many people using it yet, so you'll probably get enhanced service - I can help step you through the application, the research process.

Check it out and download for Mac. Going open source very soon.

ra-h.app