AlphaAvatar v0.6.6 shifts the framework's focus toward clearer runtime contracts and modular capabilities rather than introducing a single large feature. The release expands Environment Memory to include audio segments alongside visual observations and transitions memory updates from periodic polling to an event-driven model.
- Audio segments are now processed as derived perception observations with independent timing and context.
- Environment Memory uses adaptive accumulation triggered by relevant observations instead of timer-based polling.
- Qdrant and LanceDB backends now share a common runner protocol, decoupling storage choice from runtime behavior.
- Plugins expose capability descriptions to the Avatar context, allowing the model to dynamically understand available features.
- Perception history retention is increased and Stream/Observation schemas are made more explicit for better multi-consumer support.
- LiveKit-specific logic is isolated to entrypoints, treating it as an integration boundary rather than a core architectural dependency.
These changes aim to keep perception, memory, identity, and tools modular and observable while ensuring the model remains the central reasoning layer connecting them.