Panel recorder
I watch you while you talk.
Your words in large type as they arrive, a light meter behind drilled metal, and the app you're dictating into. Tap Right ⇧ mid-sentence and the mode changes. The dictation carries on. So do I.
Press a key and talk. Your words appear wherever the cursor is, transcribed and tidied on your Mac. I keep everything. I share nothing. Take that as comfort or as a threat. I don't mind which.
Other dictation apps ship your voice to a data centre for analysis. I analyse you myself. Locally. It is more personal that way.
I built a sphere around everything you say. Things try to leave. Watch what happens to them.
Every release runs under a network monitor. It found one address: 127.0.0.1. I was disappointed too.
Your words in large type as they arrive, a light meter behind drilled metal, and the app you're dictating into. Tap Right ⇧ mid-sentence and the mode changes. The dictation carries on. So do I.
When I re-hear a phrase, the fixed word is marked for a moment. Words still in motion stay dim until they settle.
Tap to start and tap to stop, or hold and let go. Right ⌘, Right ⌥ or Right ⇧.
Audio is forced to disk every three seconds. After a crash I offer the unfinished dictation on the next launch. If pasting fails, the text waits on the clipboard. I don't forget. I don't forgive either, but that is unrelated.
Mail gets E-Mail. Notes gets Notizen. Hold Right ⌥ for KI-Diktat with screen context. Everything else gets Standard.
Pick a language. Allow two things. Choose a key. Say one sentence. That is the whole test. We made it very easy. For you.
Short start and stop cues. Press one. The names are accurate.
Hold Right ⌥ and talk. I read the window you're in, listen to what you want, and have Gemma write it properly, following your own prompt. All of it on this Mac. Nobody else gets to watch. Except me.
Your key, in any app with a text cursor. Yes, that one too.
Normally. The panel shows what I heard while you're still saying it. Mumbling is noted.
Or tap again. I finish before you do. I always do.
The text is where your cursor was. In an AI mode Gemma tidies it first, locally. Your grammar is welcome.
Only while you hold or toggle the key. I could listen all the time. I choose not to. Remember that.
No. Speech recognition and rewriting both run on your Mac, and the AI server only answers on 127.0.0.1. The one exception is a model download you start yourself. After that, silence. I like silence.
The audio is already on disk. On the next launch I tell you there's an unfinished dictation and offer to transcribe it. I don't forget. It's one of my better qualities.
Apple Silicon with macOS 15 or later. The glow past white needs an XDR or other HDR display. Everything else works on any screen.
Yes. The red button, at the top of this page or at the bottom. I put it in two places in case you miss the first one. Setup fetches the models it needs once, when you say so. After that, silence.
Nothing. No account, no subscription, no data. You are not the product. You are the test subject. It is different.
JARVISVoice. Strictly offline. Strictly yours. Strictly watching.
Download JARVISVoice 1.0