fape
Speak naturally. Keep your language. Stay in your application.
fape is the first product I’m building at TokenSchmiede: a dictation app that turns speech into text in the application you’re already using.
Use it for a message, a note, a longer piece of writing, or a conversation with an AI agent. Start dictating with a keyboard shortcut, say what you want to say, and let fape insert the result into your focused window.
Your words, in your language
fape supports English, German, and mixed-language speech in the same workflow. Optional correction tidies spelling, punctuation, and recognition mistakes while aiming to preserve your meaning and tone. You can also use dictation without correction.
Names and technical vocabulary matter. A dictionary lets you add the terms you use and review suggested spellings. For messages, an editable draft gives you a chance to check the text before inserting it.
More than dictation
Voice actions use a separate shortcut. They can save a note, open an application, or open a website. They are a way to turn a spoken intention into a small action, while keeping ordinary dictation focused on writing text.
Where the beta stands
The current private beta is for recent Omarchy 4 desktops on x86_64 Linux. It uses Soniox for speech recognition and, optionally, Cerebras for correction and voice actions. You bring your own API keys; provider usage is billed by those services. No fape account is needed.
Speech recognition uses a cloud service, so this beta requires an internet connection. Audio is sent to Soniox; text used for correction or voice actions is sent to Cerebras. The app’s settings explain its additional options and local history controls.
Beta builds have an expiry date. Updates preserve your keys and settings. I’m looking for feedback on how fape works in real writing workflows, including the moments when it doesn’t.
Join the beta waitlist, or get in touch if you’d like to ask a question.