Iris Core

What it does

Core is the voice pipeline. It listens for the wake word, transcribes what follows as you speak, passes it to a local language model with the conversation so far, and speaks the answer back. Every stage runs on the machine; no audio or text is sent anywhere.

Speaking without gaps

Speech synthesis runs one sentence ahead of playback: while one sentence plays, the next is already being synthesised. There is no pause between sentences, and the first audio arrives as soon as the first sentence is complete, typically one to two seconds, instead of after the whole answer.

Interrupting

The stages talk over a typed event bus instead of calling each other. Speaking over Iris stops its speech immediately, and cancelling a request reaches every stage at once.

Models

Any GGUF model runs through llama.cpp, so changing the model is a configuration change, not a code change. Conversations keep multi-turn history, and the assistant's persona is a YAML profile.

Status