Architecture
Architecture
System Overview
FNSE (Fractal Neural Simulation Engine) is a production-grade, self-evolving multi-agent simulation framework built on five core pillars:
graph TB
subgraph API_Layer
REST[REST_API]
WS[WebSocket]
end
subgraph Orchestration
MacroSwarm[MacroSwarm_Orchestrator]
Scheduler[Tick_Scheduler]
end
subgraph Intelligence
GraphRAG[GraphRAG_Engine]
SkillCompiler[Skill_Compiler]
LLM[LLM_Providers]
end
subgraph Safety
Safeguards[Safeguard_System]
CircuitBreakers[Circuit_Breakers]
Rollback[Auto_Rollback]
end
subgraph State
Redis[(Redis_State)]
VectorDB[(Vector_DB)]
GraphDB[(Knowledge_Graph)]
end
REST --> MacroSwarm
WS --> MacroSwarm
MacroSwarm --> Scheduler
Scheduler --> MacroSwarm
MacroSwarm --> GraphRAG
MacroSwarm --> SkillCompiler
MacroSwarm --> Safeguards
GraphRAG --> VectorDB
GraphRAG --> GraphDB
SkillCompiler --> Redis
Safeguards --> CircuitBreakers
Safeguards --> Rollback
MacroSwarm --> Redis
LLM -.-> GraphRAG
LLM -.-> SkillCompiler
Core Components
1. MacroSwarm — Hierarchical Agent Orchestration
Manages populations of agents with specialized roles, handles tick-based execution, message passing, and cross-agent consensus.
2. GraphRAG — Graph-Based Retrieval-Augmented Generation
Combines vector similarity search with knowledge graph traversal for multi-hop reasoning and episodic memory.
3. SkillCompiler — Recursive Self-Improving Skills
Dynamic code generation, sandboxed execution, test-driven compilation, and skill versioning with dependency tracking.
4. SafeguardSystem — Enterprise Safety
Computational circuit breakers, divergence monitoring, automated state rollback, and multi-level alerting.
5. REST API — Full-Featured Server
FastAPI server with epoch lifecycle management, real-time WebSocket streaming, GraphRAG query interface, and skill compilation endpoints.