Hopper

Hopper serves open-source models for real-time voice agents: a low-latency LLM behind a standard chat-completions API, plus text-to-speech and speech-to-text. Hopper speaks text with an inline audio player, transcribes recordings you attach (WAV, MP3, M4A, Ogg, FLAC or WebM) with word timestamps, lists models and prices, returns tested integration code for popular voice-agent frameworks, and reviews a voice agent's prompt, tools and client code for what delays its first reply. After you sign in with your Hopper account, it shows your usage and creates API keys. Where it can run commands in your project, the plugin also sets Hopper up in a voice-agent codebase: it measures time to first token on the agent's own system prompt and tools in a 10-turn simulated call and shows the results before changing any code, and after you say go it switches only the LLM. Text and audio you send go to Hopper's API; see withhopper.com/privacy.