Voice Keyboard

Voice Keyboard is a keyboard that types what you say. Switch to it in any text field, tap the microphone, speak, and the text lands in whatever app you are using. It has no speech engine of its own and no account. Bring an API key for a supported speech service, such as Groq or Mistral, or run a server of your own. Audio goes only to the endpoint you configure. Compatible Chat Completions services can also transcribe audio; the selected model must accept your device's recording format. Voice input

• Works in any text field, in any app
• Processing queue: start the next recording while the previous one is still being transcribed
• Offline-proof: a recording that cannot be sent is kept on the device and resent automatically once the connection returns
• Several dictation languages on one key: tap to cycle, long-press to pick from the list
• Custom vocabulary for names and technical terms that get misrecognised
• Can be the voice input of another keyboard, such as HeliBoard, and hand back to it once the text is typed

• Fix punctuation, spelling and filler words
• Shorten, add emoji, rewrite as verse, or translate
• Optional recent dictation context helps recover misheard names and terms. Off by default; remembers up to 5 of your own recent dictations only in memory and sends them to the selected provider as reference. Context is shared across chats in the same app and clears when input moves to another app. It does not read conversations.
• Terminal mode turns dictation into shell commands
• Works with OpenAI, Anthropic, or any OpenAI-compatible provider such as OpenRouter; every prompt is editable

• One tap takes you back to the keyboard you came from
• Punctuation keys, accelerating backspace, send key, persistent clipboard bar
• The Enter key runs the field's own action: Search, Go, Send, Next or Done
• The microphone is off in password fields
• Spacing around dictated text follows the text already in the field
• Light, dark and automatic themes; 17 interface languages

• No analytics, no advertising, no tracking, no account, no servers of our own
• API keys and settings stay on the device and are excluded from cloud backup
• Audio is sent only to the endpoint you configured, authenticated with your own key

Optional recent dictation context helps post-processing recover misheard names and terms when earlier wording strongly matches. Off by default; sends up to 5 of your own recent dictations to your chosen provider. Kept only in memory, shared across chats in the same app, and cleared on app/private-field changes, disabling, 30 minutes idle or restart. No conversations are read. Terminal mode does not use or add context.