character#

CLI-first prototype for a headless chat-reactive VTuber that connects to stream.place.

Goal#

Take chat messages from stream.place, generate spoken replies, and drive a VTuber/avatar output stream with:

• chat intake
• LLM response generation
• TTS audio output
• avatar motion/expression events
• video/audio composition suitable for live streaming

Architecture sketch#

stream.place chat (Jetstream / ATProto)
  -> bot runtime
  -> reply planner
  -> TTS provider
  -> avatar controller
  -> ffmpeg compositor / scene renderer
  -> WHIP/RTMP uplink

Current direction#

This repo starts with a headless-friendly prototype:

• TypeScript runtime via bun
• stream.place chat adapter stub
• Gemini Live-compatible TTS/voice adapter interface
• avatar event bus for motion/expression cues
• local dev scripts for simulation and testing

Open questions#

• best headless avatar renderer on Linux
• whether to use browser-based canvas capture or native renderer
• whether Gemini Live should handle both reply generation and speech output
• whether stream.place posting should use ATProto OAuth directly or a simpler bot identity path

OpenVT path#

This repo now includes a simple OpenVT-oriented setup path:

• scripts/openvt-setup.sh clones upstream OpenVT and initializes submodules
• docs/openvt-setup.md explains the constraints and manual steps
• scripts/openvt-run-notes.sh prints the expected local streaming workflow

Run:

chmod +x scripts/openvt-setup.sh scripts/openvt-run-notes.sh
./scripts/openvt-setup.sh
./scripts/openvt-run-notes.sh

Wayland / WayVNC launch#

If the machine already has a remote Wayland session, launch OpenVT into that session with:

devenv shell -- ./scripts/openvt-wayland-launch.sh

See docs/wayvnc-openvt.md.