Now I just hold a key (Fn for me), talk, let go, and the text gets pasted wherever the cursor is, mostly Claude Code in the terminal. The default prompt part became "Modes" where you write the instructions once (clean it up, translate to English, etc) and the transcript goes through an OpenAI model before it's pasted.
It uses your own OpenAI key and calls their API directly, no server in between. My daily spend is usually around US$ 0.15.
It's cloud only, so if you want local models Handy or VoiceInk are better picks. No signed release yet, you build it with Xcode using one script. MIT licensed.