For years I paid for dictation. Superwhisper, Wispr Flow, a few subscription apps for turning voice into text. A few weeks ago I dropped all of them for an open source project built by one guy: FreeFlow.
I write prompts all day now. Agentic coding is mostly that, and what comes out depends on what goes in. Typing a good prompt takes me longer than saying it, so dictation became part of how I work.
It uses an API, not a local model
FreeFlow is free and MIT licensed, with no paid tier behind it. It does not run a local model. It calls an API, Groq by default, running Whisper. That is the part most people object to, because of privacy and because it needs a connection. For me it works the other way: I still have Intel Macs where local models barely run, and an API gives me a good model on hardware everyone already wrote off. Groq is fast.
Edit mode
You select text you already wrote, hit the shortcut again, and instead of dictating over it you ask for a change. Shorten this. Make it bullets. More formal. It rewrites the selection without my hands leaving the keyboard.
What I gave up
The paid apps had more features. What I use is transcription and editing, so most of those did not matter to me. The one I miss is modes: I had one for editing in English and another for Portuguese. In FreeFlow I speak in the language I want the text to come out in. Free and open source matters more to me than that.