Launched this week
SpeakoFlow
Free, open-source voice dictation and AI assistant, offline
113 followers
Free, open-source voice dictation and AI assistant, offline
113 followers
SpeakoFlow puts your voice over your whole desktop. Speak, and your words land in any app — email, editor, chat, terminal. Say "Hey Flow" and it writes the whole reply from what's on your screen. Ask the assistant about what you're looking at and hear the answer back. It also cleans up your dictation, translates as you speak, and learns how you work. Everything can run on your machine — speech-to-text always does. Free, open source, MIT. Windows, macOS, Linux.
SpeakoFlow
Hey ! I wanted to try the app on macos but couldn't download it through your website. it says the file is damaged.
SpeakoFlow
@enzo20 Thanks for flagging it, and good news: your download is fine!
macOS just asks for one extra confirmation the first time you open an app that isn't
notarized by Apple yet. It's a one-time step per version, takes about ten seconds, and
the exact instructions are here: (https://www.speakoflow.com/docs#install-macos)
On the notarization: Apple's developer program is a paid yearly subscription, and
SpeakoFlow is a free open-source project I build solo. It's planned, just not done yet.
Once it's set up, macOS opens it like any other app and this step disappears completely.
Really appreciate you taking the time to try it. Let me know how you get on!
@abhishekbarali Hey got it ! Thanks for the quick reply. I just tried the app briefly :
Very interesting app. It feels remarkably cohesive and well-thought-out in its design and execution.
One friction point I noticed with the dictation feature: consecutive dictations don't seem to register seamlessly. After completing one dictation, initiating a second one requires an intermediate tap on the screen rather than flowing directly into the next input.
A potential refinement here : perhaps the transcribed text could persist on screen for a few seconds when no active text field is selected. As it stands, if the dictation isn't immediately routed somewhere, the output is effectively lost unless the user navigates to the history to retrieve it.
Lastly, the assistant feature appears to be fairly battery-intensive. Some optimization there could meaningfully improve the overall user experience, especially for extended use.
Let me know if you've already given thought to these points. it's entirely possible you've already addressed them and I simply overlooked it.
SpeakoFlow
@enzo20 Thanks for actually trying it and writing this out.
2. Can't reproduce it on Windows, back to back dictations fire
instantly for me. My guess is transcription time. The first run after
launch has to load the model, and while one is still processing the
hotkey gets ignored so it doesn't cut off the one in flight. On my
machine that's over instantly so I never see it. There could be more to
it though, so if you open a GitHub issue with your OS I'll look into it
properly.
3. That's already covered. Settings > Dictation > Clipboard handling >
Copy to clipboard, and every transcription lands on your clipboard so
you can paste it whenever. Copy and paste felt common enough that I
didn't put an on screen indicator behind it, I wanted to keep things
clutter free 😅. But I do like your idea, I'll think about it.
4. Yeah, local models are heavy, and that's mostly the model rather
than the app. Two things that help. There's a smaller option in the
model picker, Gemma 4 E2B, which is much lighter and fine for simple
stuff, just expect less from it on anything complex. And what I do
personally is run the assistant on an API key on my laptop so battery
lasts, and fully local on my desktop where power isn't a concern. Small
models are improving fast too and I add new ones as they land, so this
side gets better on its own.
I use this daily so it's getting worked on constantly. Anything else,
open an issue and I'll get to it.
interesting
SpeakoFlow
@madalina_barbu Thank you! Happy to answer any questions about how it works.