Aurora
Voice + conversational AI
Talk to an Aithos character by text or voice. Aurora combines a character prompt, conversational AI, speech recognition and spoken replies.
What you can do
- Send a text message and read a character-based response.
- Record up to 30 seconds of speech, transcribe it, and hear an AI-generated voice reply.
- Explore how a character’s traits and narrative shape its conversational style.
Talk to Aurora
The interactive demo requires JavaScript and available backend services. The implementation, sources and limitations remain readable below.
Why it exists
Aurora explores the transition from a visual character to an interactive experience. Its humility, friendly expression, Patience Brooch, Laser Tail Ponytail and underwater setting come from Aithos Humility Aurora 290ba.
How it works
Text → Express /chat → OpenAI GPT-4o → reply. Voice → browser WAV recording → /recognize → Whisper transcription → the same chat flow. Optional spoken replies use /speak → TTS-1 with the nova voice.
Architecture and stack
React manages the visible conversation. RecordRTC StereoAudioRecorder produces WAV audio, including on Safari. Express isolates provider credentials; Redis enforces shared usage limits. The model receives the character instructions and your current message only.
Technical challenges
Browser microphone permissions, Safari recording formats, speech playback restrictions and asynchronous requests all affect a voice experience. The implementation uses WAV recording, releases microphone tracks, limits recording duration and keeps text replies available if audio fails.
Privacy and safety
Messages, recordings and reply text are sent through this server to OpenAI for response generation, transcription or speech. Audio uploads are processed in memory. Do not share sensitive information. Aurora is a fictional AI character, not a therapist or a clinical service.
Limitations
Responses can be wrong. The displayed conversation lasts only in this page; prior messages are not sent as model context. There is no persistent character memory or wallet verification. Usage limits and service availability may interrupt a session.
Implementation lessons
A character prompt provides a starting point for personality, but the surrounding system must also explain memory boundaries, manage microphone resources and handle provider failures. Persistent identity requires a separate data and consent design.
Next work
Persistent AI character behavior is research. Memory, ownership-aware access and long-term character development are not shipped features of this demo.