chappie-architecture

name	chappie-architecture
description	CHAPPiE's brain, memory, and life simulation architecture. Use when modifying core pipelines, brain agents, steering, or global workspace.

CHAPPiE Architecture

CHAPPiE uses a brain metaphor across three interacting layers: Brain, Memory, Life.

Brain Pipeline (brain/)

Agent	File	Purpose
Sensory Cortex	brain/agents/sensory_cortex.py	Input classification
Amygdala	brain/agents/amygdala.py	Emotional weighting
Hippocampus	brain/agents/hippocampus.py	Memory retrieval & encoding
Prefrontal Cortex	brain/agents/prefrontal_cortex.py	Response strategy & tone
Steering Manager	brain/agents/steering_manager.py	VAD mapping, alpha, composite modes, layer editing
Global Workspace	brain/global_workspace.py	7 signals with salience bundling
Brain Pipeline	brain/brain_pipeline.py	Orchestrates all agents

Key Integration Points

• web_infrastructure/backend_wrapper.py → Main CHAPPiE backend class
• api/main.py → FastAPI app entry
• brain/__init__.py → get_brain() factory

Memory (memory/)

Component	File	Purpose
Memory Engine	memory/memory_engine.py	Episodic search, vector DB (ChromaDB)
Sleep Phase	memory/sleep_phase.py	Consolidation, replay, compression
Forgetting Curve	memory/forgetting_curve.py	Ebbinghaus decay model
Context Files	memory/context_files.py	soul.md, user.md, preferences
Short-Term	memory/short_term_memory.py	JSON-based STM with timestamps

Life Simulation (life/)

Component	Purpose
Service	Prepare/finalize turns, homeostasis
Goal Engine	Goal competition
Planning Engine	Long-term planning
Social Arc	Relationship/attachment models

Key Rules

• chappie-training.service must start training_daemon, never training_loop
• Absolute paths in systemd ExecStart and WorkingDirectory
• Steering: local Qwen first, vLLM preferred, Cloud APIs as fallback
• brain/__init__.py triggert ALL brain imports (ollama, cerebras, vllm) — test imports müssen das mocken