🔊 Hear Llama 3 Through Kokoro

Type a short message. Llama 3 8B answers. Kokoro speaks the answer out loud through your speakers. All running on rentahal.com — one RTX 4060 in Newburgh, NY.

Type something

0 / 100 characters
connecting to rentahal.com…
Waiting. Click "Ask & Speak" once connected.

What just happened

  1. This page opened a WebSocket to wss://rentahal.com/ws
  2. You typed text. We sent {type:"prompt_text", text:"…"}
  3. The realm classified the intent, routed to Llama 3 8B Instruct running locally on a 4060
  4. Llama generated an answer (you saw the text)
  5. The realm sent the answer to the Kokoro GPU TTS microservice (port 9998 on the realm box)
  6. Kokoro synthesized WAV audio chunks; the realm streamed them back as bus_tts_chunk events with base64-encoded audio
  7. This page decoded them and played through your speakers

No API keys. No cloud LLM. No third-party TTS service. The entire pipeline runs on one household GPU. You just used the same WebSocket protocol the real RENT-A-HAL UI uses.