There's a few things:
1. This is significantly more efficient than most other apps. It was built using Rust and GPUI, so the app itself runs in ~5MB of memory. Most other apps are built using Electron or Tauri which results in a lot more memory and CPU usage.
2. This uses different backends for different systems, with the macOS backend being built on top of WhisperKit. This further adds to the efficiency.
3. AI enhancement is supported (but not required) using APIs, but also your local Claude Code or Codex CLI. That means for a large percentage of people AI enhancement requires basically no setup.
4. I am obviously highly biased, but I think this both looks and feels better than any other open source solution I've tried. This is of course going to be up to personal preference.
There's other things I could mention, but those are the big ones in my mind. I'm building this to take on WisprFlow, so the whole goal is to do what they do, but better and FOSS.
I've switched almost entirely to voice dictation on my phone using a local model, and I really want to do something similar on my desktop. I've been using Handy on Linux, but it's quite slow, even though I have a GPU available. I'll give Sayso a spin, thanks for sharing! I'll report back with any interesting findings.
Having recently tried Handy (on Wayland/Gnome) it seemed very slow.
On first look this seems very slick and the default settings are responsive on my aging laptop.
9 comments
[ 0.32 ms ] story [ 10.0 ms ] threadThere's other things I could mention, but those are the big ones in my mind. I'm building this to take on WisprFlow, so the whole goal is to do what they do, but better and FOSS.
Yet, with so many dictation apps, it makes me wonder how many are honeypots or something.
I'm very impressed so far.