OhmsVoice.
Speak, and get clean text on the clipboard.
Nothing records until it can transcribe.
The first time you open OhmsVoice, it asks how to listen: the on-device speech recognition built into your Mac, iPhone or iPad, where it supports your language, or an open-source speech model downloaded once with your consent (483 MB, 25 languages), pinned to a verified revision, checksummed, and load-tested before it is marked ready. The microphone stays closed until one of them is ready. Either way the audio stays on the device and is deleted as soon as the text is ready. There is no cloud fallback.
Rewrites go only where you send them.
Raw Transcript and Clean (On-Device) never call a provider. Cloud presets stay blocked until you save a key and explicitly allow the named provider. A rewrite then sends the transcript, the preset's instructions and any spellings you saved directly to that provider, never audio. Through version 1.1.2, rewrites started on a Mac also include the name of the app you were dictating into; later versions never send it. On a Mac you can point rewrites at a model server running on that same Mac instead. Deleting a key revokes that provider's approval. Nothing is proxied through Ohm.
Keys live in the Keychain.
API keys are pasted, saved, and deleted in Settings, and stored in the system Keychain under com.silosystems.ohmsvoice. There is no Ohm account, no Ohm server, no telemetry, and no analytics.
On a Mac, hold a system-wide shortcut (default ⌃⌥ Space, or one of four others) and speak into any app. Hold to talk, press to toggle, or keep the default hybrid: hold for a quick phrase, tap once to keep listening. The shortcut needs no Accessibility or Input Monitoring permission, and if another app already owns it, OhmsVoice tells you. You can also record from the menu bar item or with ⌘R, and on iPhone and iPad from the Record tab. The finished text is copied to the clipboard, ready to paste wherever you were typing.
Clean (On-Device) and Raw Transcript need no key and no network. Clean removes English filler words such as “um” and “uh”, turns English spoken commands such as “new line”, “full stop” and “question mark” into layout and punctuation, and fixes capitalization. Raw Transcript is exactly what you said. Clean & Concise, Executive Email and Developer Bug / Jira Ticket rewrite the text with a provider you choose. For meeting notes, use Silo Meet.
Say a short trigger to drop in text you saved, such as an address, a sign-off or a link. Save the names and jargon you use and they come out spelled your way, even when spoken apart: “swift ui” becomes “SwiftUI”. You add, see and remove both in Settings, they are applied on the device to Clean and rewrite results, and Raw Transcript is never changed. Rewrite presets also pass your saved spellings to your provider as a hint.
Connect a cloud provider with your own API key, or on a Mac a model server running on that Mac. Cloud sending needs explicit approval for each provider, and keys stay in the Keychain. Saving a key makes that provider the active one, and it stays active across launches. Deleting a key revokes its stored approval. You pay the provider directly; OhmsVoice adds nothing on top.
On a Mac, pick the preset each application gets: Raw Transcript in your terminal, Clean (On-Device) in Messages, Executive Email in Mail. OhmsVoice follows whichever app is in front when you start speaking. Transcript history is yours to set to Off, 1 day, 7 days, or Forever, with one-tap clear. Cancel works from any state, and a result you cancelled never reaches the clipboard later.
More Ohm apps that talk only to the provider you choose.
3 related apps- Download on the Mac App Storelocalization · $39.99OhmsPolyglot
An editor and linter for Xcode String Catalogs.
- Download on the Mac App Storevision · $24.99Ohms Lens
Screenshot to design tokens, worked out on this Mac with no API key.
- View on the App Storeproductivity · $29.99Ohms Explain
Ask about anything you can see.