Redcast Cloud transcription and modes
Choose where audio is sent, add vocabulary, and rewrite speech before insertion.
Cloud transcription is an optional engine for the existing Dictation shortcut. Instead of the local recognizer, your approved audio-capable model transcribes the completed recording and the result is inserted into the field you were using.
Choose a connection
In Settings → Dictation, enable Cloud transcription and choose a model. The confirmation names the connection and model that will receive audio. Redcast supports audio-capable Gemini, OpenRouter and OpenAI-compatible routes; Anthropic is not an audio route here.
A connection using an OpenAI-shaped protocol is not proof that every model behind it accepts audio. Choose a model actually supporting that input. If the saved connection is removed or disabled, Redcast asks you to choose again instead of sending the recording somewhere else.
Languages and vocabulary
Set your languages and custom spellings for names, project terms or abbreviations. The transcription prompt handles spoken punctuation, lists and corrections. Speak a short sample containing your unusual words to check the result before relying on it for a long message.
Dictation modes
Modes change what is typed after transcription. Built-in choices include Polish, Translate to English and As a message; you can define your own instructions.
A mode can be associated with an app. The capsule names the current mode, and Tab switches modes while dictating. Check it before finishing, especially when moving between a chat app, an editor and a browser.
On a model that can hear and rewrite in one request, the mode can be handled together. A transcription-only route may need a second request to rewrite the transcript, which can add latency and provider usage.
Context from the app in front
Use the app in front can provide the app name and, when permitted, additional window context as a hint for the transcript. A screenshot requires the separate Screen Recording permission and context setting. This context belongs to that request and is not saved in Dictation History.
A title or screenshot may contain private text. Enable the level you actually want to send to the approved provider.
Audio and failure recovery
The network request happens after recording ends. Silence is trimmed and a recording with no detected speech is not sent. Successful recordings are discarded from memory after use.
A failed transcription can keep its audio locally for retry. Retrying with a model on another connection asks again before sending there. Retry updates the history entry; it does not type into whichever app happens to be open.
Costs
Cloud transcription and any rewrite request consume the selected provider's allowance. AI Usage records reported tokens and cost where available. On-device transcription remains available when you want recognition without a cloud request.