Implemented
- Local-first voice runtime with wakeword, STT, response generation, TTS, and runtime state guards.
- Wakeword microphone ownership by runtime mode to reduce backlog and STT conflicts.
- Read-only WebUI observability with page-based navigation.
- SQLite repositories for users, memories, conversations, events, logs, faces, commands, and system state.
- Camera initialization and frame capture through
CameraManager. - Vision auto-capture for periodic snapshot and analysis.
- Vision quality analyzer with brightness, contrast, exposure, blur, backlight, image quality, and face readiness metrics.
- Face detection with Haar Cascades as an implemented experimental baseline.
- WebUI vision calibration panel with latest snapshot, quality metrics, face detection counters, reject reason, and bounding boxes.
Next
- Improve detector stability across lighting, distance, blur, pose, and false positives.
- Add YuNet/OpenCV DNN backend as a stronger detector option.
- Add temporal smoothing and tracking for face detections.
- Add face enrollment flow.
- Add recognition embeddings and identity matching.
- Integrate
FaceRepositorywith the enrollment flow. - Tune vision thresholds and validation guidance from real camera sessions.
Pending
- REST API for external clients beyond the current read-only WebUI endpoints.
- Command-line interface.
- WebSocket protocol for real-time bidirectional communication.
- Planner/task system with goal-driven task decomposition and scheduling.
- Robotic abstraction layer for motion, sensors, and physical-world state.
Future
- Multi-user recognition.
- Per-user memory/personality.
- Semantic similarity search with embeddings.
- Long-term memory consolidation and importance decay.
- Robotic motion abstraction.
- Planner/task system integrated with voice and memory.