While working on my last startup I found myself spending most of the day with AI tools. I felt typing slowed me down. Since I was working on voice AI, I built a dictation tool for myself. It quickly became part of my workflow, and when I shared it with other developers they also loved it. That became Voibe.
Voibe is a fast dictation app for Mac that runs entirely on-device. It uses quantized Whisper models locally, so everything stays private and works without internet. It transcribes in real time and works in any macOS application.
A couple of things make it different:
- Deep integration with coding agents like Cursor and Windsurf: you can speak file and folder names and it resolves them correctly from the active workspace.
- Custom dictionary for your own vocabulary of technical terms, acronyms, or product names.
Why it matters: In our workflow we found talking > typing for AI work. We speak and type differently; speech preserves intent and usually produces longer, more clearer prompts. Early users report better first-pass results when they talk to AI.
Questions for HN:
- Have you tried talking (instead of typing) to AI what's been your experience?
- Do you use dictation in dev/AI workflows today? What blocks you (latency, accuracy on technical terms, punctuation, noise)?
- For developers: would deep integration with Cursor/Windsurf—resolving spoken file/folder names from the active workspace—be useful? In which tasks (prompting, refactors, code search, commit messages)?
- Does fully offline/on-device matter to you vs cloud tools? Why?
balamuruganb•1h ago
While working on my last startup I found myself spending most of the day with AI tools. I felt typing slowed me down. Since I was working on voice AI, I built a dictation tool for myself. It quickly became part of my workflow, and when I shared it with other developers they also loved it. That became Voibe.
Voibe is a fast dictation app for Mac that runs entirely on-device. It uses quantized Whisper models locally, so everything stays private and works without internet. It transcribes in real time and works in any macOS application.
A couple of things make it different: - Deep integration with coding agents like Cursor and Windsurf: you can speak file and folder names and it resolves them correctly from the active workspace. - Custom dictionary for your own vocabulary of technical terms, acronyms, or product names.
Why it matters: In our workflow we found talking > typing for AI work. We speak and type differently; speech preserves intent and usually produces longer, more clearer prompts. Early users report better first-pass results when they talk to AI.
Questions for HN: - Have you tried talking (instead of typing) to AI what's been your experience? - Do you use dictation in dev/AI workflows today? What blocks you (latency, accuracy on technical terms, punctuation, noise)? - For developers: would deep integration with Cursor/Windsurf—resolving spoken file/folder names from the active workspace—be useful? In which tasks (prompting, refactors, code search, commit messages)? - Does fully offline/on-device matter to you vs cloud tools? Why?
Happy to answer questions here.