AriaType is an open-source voice-to-text app for macOS that lets you speak naturally and type anywhere with a hotkey. It works right at your cursor, supports local-first processing for better privacy, and helps you turn speech into polished text without breaking your flow.
I'm a CS developer with 10 years of experience. This Spring Festival, I was working on a web coding project.
About 3 days in, I realized something: The biggest challenge wasn't the cost of AI subscriptions. It was my stamina.
The constant cycle of writing prompts, correcting AI outputs, and context-switching was exhausting. So I built myself a voice input tool.
Why build it myself?
I looked at Typeless. The subscription was steep — almost matching my AI plan. I wasn't willing to pay that much for something I could build for myself. (But it's great, to be frankly)
Two hours. MVP.
LLaMA CPP + Whisper. Rough around the edges, but functional. Good enough for personal use.
Then the real work started:
After sharing it with colleagues post-holiday, I got real feedback. Some of it stung, but all of it was valuable. So I spent the following weekends polishing it.
What AriaType 0.1 does:
- Local STT models — whisper-based. Vesper for English, Sense Voice (Alibaba) for CJK - AI Polish — local small models for grammar correction, filler removal - Cloud mode — bring your own AI subscription. No separate payment - Noise reduction & silence detection — skips silent chunks to save costs - 100+ languages - Privacy by default — voice data never leaves your machine
What I'm proud of:
- 100% open source - Everything runs locally by default - Works in any app via global hotkey - Minimal footprint on Apple Silicon
The honest challenge:
The hardest part wasn't the MVP. It was the 80% after — making AI reliably modify and extend a growing codebase. Called "harness engineering."
Report
@joe_he Waiting for a windows version Man! Kudos on the Launch!
Report
Maker
@nayan_surya98 Thank you so much for your support and encouragement
Report
The works in any app promise is huge .That's where most voice tools fall short .
Report
Maker
@conrad_foster To ensure it works properly across all applications, I've tried implementing multiple fallback solutions. This has indeed involved a considerable amount of work, but I believe it's worthwhile for a better user experience
Looks good, but your homepage needs another look. It has a bunch of unfinished AI prompts describing the marketing copy you want on the page, not the actual behavior of the app.
Report
Maker
@brianellin Really appreciate, I'll try to find a better solution in next version
Replies
TL;DR
👋 Hey Product Hunt!
The origin story:
I'm a CS developer with 10 years of experience. This Spring Festival, I was working on a web coding project.
About 3 days in, I realized something: The biggest challenge wasn't the cost of AI subscriptions. It was my stamina.
The constant cycle of writing prompts, correcting AI outputs, and context-switching was exhausting. So I built myself a voice input tool.
Why build it myself?
I looked at Typeless. The subscription was steep — almost matching my AI plan. I wasn't willing to pay that much for something I could build for myself. (But it's great, to be frankly)
Two hours. MVP.
LLaMA CPP + Whisper. Rough around the edges, but functional. Good enough for personal use.
Then the real work started:
After sharing it with colleagues post-holiday, I got real feedback. Some of it stung, but all of it was valuable. So I spent the following weekends polishing it.
What AriaType 0.1 does:
- Local STT models — whisper-based. Vesper for English, Sense Voice (Alibaba) for CJK
- AI Polish — local small models for grammar correction, filler removal
- Cloud mode — bring your own AI subscription. No separate payment
- Noise reduction & silence detection — skips silent chunks to save costs
- 100+ languages
- Privacy by default — voice data never leaves your machine
What I'm proud of:
- 100% open source
- Everything runs locally by default
- Works in any app via global hotkey
- Minimal footprint on Apple Silicon
The honest challenge:
The hardest part wasn't the MVP. It was the 80% after — making AI reliably modify and extend a growing codebase. Called "harness engineering."
@joe_he Waiting for a windows version Man! Kudos on the Launch!
@nayan_surya98 Thank you so much for your support and encouragement
The works in any app promise is huge .That's where most voice tools fall short .
@conrad_foster To ensure it works properly across all applications, I've tried implementing multiple fallback solutions. This has indeed involved a considerable amount of work, but I believe it's worthwhile for a better user experience
Medium
@brianellin Really appreciate, I'll try to find a better solution in next version