The big idea
Most "AI chat" UIs are still text boxes with a microphone button bolted on. Aura Assist starts from the opposite assumption: the conversation is the interface.
Architecture: thin client, shared agent
- Voice-first React shell — UI built around three states: listening, thinking, speaking.
- Streaming pipeline — Web Audio API +
capacitor-voice-recorder, streamed to the Universal AI Agent over WebSockets, played back as it arrives. - Shared agent core — consumes the Universal AI Agent backend; provider routing, auth, rate limiting, and audit logging come for free.
- Lightweight state — Zustand + React Query + shadcn/ui.
Innovative features
- Three-state visual grammar — animated waveform encodes listening/thinking/speaking via amplitude, colour, and motion.
- Streamed playback — audio chunks play as they arrive.
- Provider-agnostic via the Universal AI Agent.
- Reusable voice primitives designed to be lifted into ClearedMind for guided meditation prompts and voice journaling.
Outcome
- A working voice chat surface usable from any modern browser, with sub-second perceived response times.
- Validated the Universal AI Agent against a real-time, multimodal client.
- Reusable voice-UI primitives ready for ClearedMind.