I got so tired of typing that I built a voice input tool. Here's what changed.

I'm a developer. I write a lot. Messages, docs, comments, emails, AI prompts — all day long.
At some point I did the math:

Average typing speed: ~60 words per minute
Average speaking speed: ~150 words per minute

That's 2.5x more output for the same amount of thinking.
And yet I was sitting there, pecking at my keyboard like it's 1995.
So I built something about it.

The problem with existing voice tools
I didn't want to build this. I went looking for something ready-made first.
What I needed:

Text appears wherever my cursor is — any app, any window
Works offline — no sending audio to some server
Free, no subscription
No button rituals — just talk, text appears

What I found: cloud-only tools, paid subscriptions, apps that only worked in their own tiny window (great, very useful, love copying text manually).
So: built it myself. Classic developer move.

VoxBee — voice input that actually fits into your workflow
It's a free, open-source Windows app. Runs locally on whisper.cpp. No internet needed after setup. No data leaves your machine.
The key feature: auto mode.
You open any text field. You start talking. Text appears. You stop talking. It stops.
No buttons. No "click to record." No switching apps.
The workflow difference is real — once you dictate a few documents or long messages this way, going back to typing feels genuinely slow. Your thoughts move faster than your fingers. Now your output can too.

The productivity gains that surprised me

1. AI prompts get better When I type prompts for ChatGPT or Claude, I keep them short because typing is effort. When I dictate, I give full context, more detail, better instructions. The responses got noticeably better just because I stopped being lazy with my prompts.
2. "Okay" sends the message I added a voice command: say "okay" → hits Enter. Sounds trivial. In practice, when you're dictating 20+ messages a day, never reaching for the keyboard is a genuine quality-of-life upgrade. Other commands:

save → Ctrl+S
undo → Ctrl+Z
click → mouse click
up / down / left / right → moves cursor

You can add custom ones too. I have a command that opens my task manager. Another that saves and closes a file. Whatever you repeat 10x a day — automate it with your voice.

1. Notes and drafts flow differently When I type, I edit as I go. Delete, rephrase, backspace. It's slow and it fragments thinking. When I dictate, I just... talk. The draft comes out messier but faster and often more natural. Then one editing pass. Total time: shorter.

Heads up: it's not magic
In auto mode, background noise can sometimes trigger transcription. Fan noise, street sounds, a door slamming — Whisper might try to transcribe it. Noise suppression helps a lot.
Also: it's a beta. Tested on my machine. Works great for me, might have quirks on yours. Open source, so if something's off — the code is there.

Performance

NVIDIA GPU (CUDA): essentially instant. You finish a sentence, the text is already there.
CPU only: 1-2 second delay with the base model. Totally fine for most use cases.
AMD: separate Vulkan build in progress, not ready yet.

Models (start with base)
ModelSizeNotestiny75 MBFast, rougher accuracybase148 MBBest starting pointsmall465 MBBetter accuracylarge-v3-turbo1.6 GBBest if you have a GPU

Who this actually helps
Great fit if you:

Write long-form content (articles, docs, reports)
Send a lot of messages or emails daily
Use AI tools heavily and write detailed prompts
Have wrist/hand pain from typing
Think faster than you type (most people do)

Probably not worth it if you write 50 words a day and are happy with that.

Setup (one time, 3 steps)

Download installer → Releases
Grab model ggml-base.bin → Hugging Face
Get engine whisper-blas-bin-x64.zip → whisper.cpp releases

Put the models/, cpu/, gpu/ folders next to the installer. Done.
Windows 10/11, any mic, 500 MB disk space.

Repo: github.com/boris-agent007/Voxbee
If you try it, let me know how it goes. Genuinely curious whether the productivity angle lands for people outside my own workflow.
— Secret Agent 007 🕵️

P.S. Dictated this post lying on my couch. Then had an AI clean up the draft. A post about voice productivity, written via voice, edited by AI. Make of that what you will.