Voice pipeline
Desktop edge voice and browser iPhone voice share the brain but use different audio paths.
Desktop edge
The edge integrates wake-word detection, recognition, turn handling and synthesis. Provider choices live in Settings. Installed extras and models determine which local engines can be built.
Barge-in and duplex behavior depend on the microphone, speaker and endpoint. Test real hardware; configured latency targets are not measured guarantees.
Browser gateway
The iPhone path accepts a bounded recorded utterance, transcribes it, submits text to the brain and streams reply text with sentence audio. It does not transcribe each character while recording.
The orb uses published lifecycle and playback state. User-facing progress can appear in dialogue; private internal reasoning is not exposed as a transcript.
Implementation sources
Repository paths below describe the implementation. They do not certify a running deployment or a configured external account.
src/mushegh/edge/assistant.pysrc/mushegh/edge/stt.pysrc/mushegh/edge/tts.pysrc/mushegh/brain/hud_gateway.py
Subsystem directory · HUD guide · Verification and known limits