package:a5c-ai-voice-adapter
Media Governance Bridge (audio + video) overview
PROPOSED (not yet built). Thin TypeScript bridge letting a realtime media agent call babysitter-governed tools over MCP — across both the voice path and the animated-avatar VIDEO path on kradle/Jitsi. Cosmetic animation (expression/ posture/gaze/visemes) runs on a realtime fast path; consequential visual actions (canvas content, screen-share/VNC, external video metadata) are gated like any sensitive tool via in-process babysitter runs (createRun -> orchestrateIteration -> commitEffectResult) with policy + auth./destroy. breakpoint approval gates and a replayable journal — all OUT of the audio/animation hot path. Extends the existing kradle CRDs (AgentStack.jitsiConfig, JitsiMeeting, AgentAppearance/ AgentVoiceProfile) rather than minting a parallel CRD. See docs/research/voice-governance-bridge-spec.md + realtime-agent-gaps.md.
Attributes
- mcp-server
- crd-controller
- channel-bridge
- webrtc
- sip