Voice and read-aloud

You can talk to kral and it can talk back. Both directions run through the same gateway as everything else, billed against the same credit.

Dictating a prompt

Click the microphone in the prompt box, speak, click again to stop. The recording is transcribed by a speech-to-text model (Whisper-grade) and lands in the input as text, so you can correct it before sending. The browser will ask for microphone permission the first time.

Dictation handles natural speech well: punctuation, multiple languages, technical vocabulary. Transcription is billed per second of audio; a short dictation costs a fraction of a cent.

Read-aloud

Every answer has a read-aloud control. The text is converted by a text-to-speech model with natural voices and billed per character. Useful for long answers while you do something else, or for proofreading your own drafts by ear.

Voice settings

In settings, the speech tab lets you pick the voice for read-aloud and tune dictation behavior, like automatically sending after you stop talking.

API access

The same speech models are available programmatically via /v1/audio/transcriptions and /v1/audio/speech. See Endpoints.