Rime launches Prism for speech-to-speech conversations

An inspectable checkpoint before speech

Rime launched Prism on October 28 and offered a free trial. It describes a system that listens to audio, writes its intended reply and continues hearing the user while speaking, allowing application rules to inspect text before audio is delivered. These are Rime’s design claims, rather than independently established task performance. [1][2]

The launch post says self-serve reinforcement-learning and supervised-fine-tuning APIs were being built with beta customers. That work should be distinguished from broadly released self-serve training or public model weights. Writeble likewise separated the hosted documentation from the company’s broader deployment claims. [1][2]