Conversation, voice, and companionship

Status: future idea, no implementation commitment.

Direction

SkillNet may incorporate several features related to audio and companionship. They must be kept separate so they can be improved or replaced without coupling the whole application.

Audio in chat

The person can send an audio message to the chat. SkillNet transcribes it and the chat responds in text.

audio -> transcription -> chat -> text response

This does not imply the chat responds with voice, nor that a live conversation is started.

Realtime conversation

Realtime is a distinct feature for holding a live voice conversation. Its integration must sit behind an abstraction so as not to couple the rest of SkillNet to GPT Realtime or to any particular provider.

Mascot

The mascot visually represents the learning companion. It must not contain the chat, voice, or node-reading logic inside it. It receives simple signals from the system and its appearance can change without modifying those features.

The concrete way to improve how it interprets and accompanies nodes remains pending design and validation.

Podcasts

Podcast generation continues to be a separate feature. Both the generation studio and the output quality will be improved, without turning the podcast into part of chat or Realtime.

The modular, configurable plan is defined in podcast-studio-plan.md.

Relationship to audience modes

These features can be part of SkillNet’s common core when useful in a course. The organization mode keeps the current company/employee focus. The individual mode, if implemented in the future, will reuse the same features without employee or talent management.

This document does not yet define additional use cases, evaluation flows, memory, implementation priorities, or detailed interface behavior.