- Personality / Role module and db with persistence.
- Choose RAG document groups per personality / topic.
- General separation of document categories.
- Per User history full storage in vectordb for history context.
- Multi-LLM communication
- LLM Broker for chat like communication handling between LLMs + Chat memebers.
- Prompt optimization improvements. May need a specialized model for this.
- 3D character animation layout w/out adobe character animator.
- Video generation from images, for muxing output into a video stream.
- Text to video... someday
- Music improvements, replay old music jukebox mode, more continuity...
- Image generation context improvements.
- Header information on priority / must create image flags etc.
- Improve throttling / go over.
- Generally refactor each module with common handling of each function that is generic.
- Separate python virtual environments per module and automate maintaining them.
- Get manager and conrfiguration handler working and tested.
- Stats and Chat information etc burn-in on images for Twitch.
- Debug / information burn-in for debugging / nerd stats output.
- Possibly history context on prompt optimizer?
- RTMP output stream / or format we can send as RTMP for use as a network output tap.
- Use softsubs VTT / ASS possibly in the future with video mode / LLM help.
- Build filters in FFmpeg to do each of the model functions and connect them through ZMQ (Someday...)
- Ingescape exploration to have things distributed yet directly interact as one system (ZMQ next generation).