bandicot

/speak answers with the recording itself

/speak goes straight to the voice model, and the recording is the reply, with no chat model in between.

Product

Type /speak and some words, or /speak alone to hear the last reply. The voice model reads them at once and the player is the answer, right under your line. No chat model runs, so it costs no tokens and nothing waits on a provider.

Put a recording first, as in /speak voice.wav Hello, and the words come out in that voice.

When an agent speaks during its own work, the recording now shows under its reply too, instead of inside the work steps.