Overview
ECHO_feeling is an internally-driven emergent AI companion desktop application for Minecraft. The AI acts as an independent player character with long-term memory, remembering shared adventures with the human player.
Implementation Approach: Internal emergence — based on cognitive science architecture (7-layer cognition, PAD/OCC emotion, BDI intention), with partial hardcoding following theoretical frameworks.
Platform Scope: Windows 10/11 x64 only.
Key Differentiators
| Feature | Standard Game AI | ECHO_feeling |
|---|---|---|
| Memory | None/Single session | Cross-session long-term memory |
| Operation | Voice suggestions only | Autonomous character actions |
| Emotion | None | PAD/OCC emotion model |
| Intention | None | BDI intention system |
Architecture
Built with Electron + Vue 3, core components include:
- Mineflayer Bot Controller: Headless Minecraft client for full game interaction
- LangChain Agent: DeepSeek API integration for natural language processing
- 7-Layer Cognitive Architecture: Progressive cognition from perception to self-awareness
- Layered Memory System (L0-L3): From identity to deep retrieval
- Code Execution Sandbox: node:vm + static checks, upgradable to isolated-vm
- Local Vector Database: sql.js storage with disk encryption support
Code Agent Mode
Player command → Mineflayer listener → LangChain Agent
→ Generate JavaScript code → Local sandbox batch execution
→ Store memory → Reply to player
Key design: Single LLM call generates complete operation code, zero API consumption during execution.
Memory Layers
| Layer | Content | Load Timing |
|---|---|---|
| L0 | Character identity | Always loaded |
| L1 | Core memory summary | Every conversation |
| L2 | On-demand vector retrieval | AI-initiated |
| L3 | Deep hybrid search | AI-initiated |
Development
| Command | Description |
|---|---|
npm run dev | Development mode |
npm run build | Compile to out/ |
npm run typecheck | TypeScript check |
npm test | Unit tests |
npm run e2e | Lightweight smoke test |
npm run pack | Build Windows installer |
Status
Core functionality complete, optimization phase. Seven-layer cognitive architecture, PAD/OCC emotion system, BDI intention system, and layered memory implemented.