Runs Aqua's normal dictation pipeline and returns when the text is ready. The request stays open while Aqua transcribes and formats the audio, for up to 210 seconds. It does not create a public background job. Send an Idempotency-Key and retry the same request with the same key if the connection closes. A terminal result replays for 24 hours without another inference or charge. After Aqua dispatches the audio, an uncertain upstream failure is terminal for that key so a retry cannot create a second dictation.
Aqua applies the user's language setting, dictionary, replacements, custom instructions, casual messaging preference, and learned preferences. A request language overrides the saved language for this dictation. The returned session_id identifies the session stored in Aqua history. Privacy and zero-data-retention settings still apply: the text is returned to the caller but is not stored when retention is off.
Use operation=edit with selected_text to run Edit Mode. The optional text and app fields give Aqua the same surrounding context it uses when dictating into an app.
Successful requests are billed by decoded audio duration at $0.49 per hour, with a 10-second minimum. Dictate and Edit Mode use the same rate. Failed requests and idempotent replays are not billed again.
