MindMirror Micro-Coach
Real-time micro-coaching with OpenAI Realtime API and emotion detection
Intelligent Emotional Guidance
Minimal Node.js WebSocket client for OpenAI Realtime API that provides real-time micro-coaching based on emotional state, focus levels, and task engagement. Sends JSON state updates and validates responses against a strict contract.
Includes three operational modes: OpenAI Realtime WebSocket, offline rule-based engine requiring no API, and DeepSeek integration for normal LLM processing.
Operating Modes
Core Features
Emotion Detection
Real-time emotion analysis tracking focus, valence, arousal, and dominant emotions for personalized coaching responses.
JSON State Updates
Sends structured JSON state updates as plain text turns including language, focus, valence, arousal, emotion, time on task, and consent status.
Contract Validation
Validates assistant responses against JSON contract (message/follow_up_question/nudge_type/tip/cooldown_sec/content_warnings).
Offline Rule Engine
Deterministic rules engine that mirrors spec trigger logic, returns contract-valid JSON, and requires no API.
System Prompt Loading
Loads MindMirror system prompt from prompts/mindmirror_realtime_coach.txt for session instructions.
Multi-LLM Support
Works with OpenAI Realtime (gpt-4o-realtime-preview-2024-12-17) and DeepSeek API (deepseek-chat) with configurable models.
Technical Specifications
Ready for Real-Time Micro-Coaching?
Deploy emotion-aware coaching with OpenAI Realtime API and offline rule engine