Tian AI's Three-Layer Architecture: A Developer's Perspective

Tian AI is organized into three logical layers, each with a distinct responsibility.

Layer 1: Perception & Communication

Talker Module

• Handles conversation management
• Multi-turn context tracking
• Emotion-aware responses
• LLM-driven follow-up questions

File: talker/__init__.py

Layer 2: Reasoning & Knowledge

Thinker Module

• Three-layer reasoning (Fast/CoT/Deep)
• Query routing based on complexity
• Knowledge base integration

Knowledge Module

• SQLite FTS5 knowledge retrieval
• Million-entry concept database
• 0.04-second average query time

Files: cot_engine.py, knowledge/__init__.py

Layer 3: Action & Evolution

Agent Module

• LLM-driven task planning
• TaskQueue with dependency resolution
• Safety whitelist for autonomous execution

Self-Evolution Module

• AST-based code analysis
• Automated patch generation
• XP + leveling system

Files: agent/__init__.py, llm_agent.py, self_modify.py

Cross-Cutting Concerns

LLM Management

• llm_manager.py: Process lifecycle
• llm_bridge.py: API communication
• prompt_cache.py: Response caching

Performance

• Project: 770+ Python files, 171K+ lines
• Core: 6 modules, 3 extension languages
• Response time: 0.04s (knowledge) to 60s (deep reasoning)

Developer Notes

• Each module is independently testable
• Minimal external dependencies (just Flask, Gradio, requests)
• Designed for easy module replacement
• All modules work offline

This architecture allows Tian AI to be both powerful and maintainable.