Overview

The GNUS.ai system integrates multiple open-source technologies to build a fully decentralized Retrieval-Augmented Generation (RAG) architecture. It leverages:

• RocksDB over IPFS for distributed storage and CRDT-based synchronization.
• Vulkan shaders and ggml for high-performance inference on GPUs.
• MNN (Mobile Neural Network) to dynamically load and execute models at the edge nodes.
• Pub/Sub communication over libP2P for distributed query handling across nodes.
• Federated learning principles to allow local model retraining and fine-tuning on each node.

This architecture ensures scalability, fault tolerance, and security while minimizing latency in both data access and inference.

graph TD;
    SystemArchitecture -->|Data Storage| RocksDB[Edge Nodes using RocksDB + IPFS]
    SystemArchitecture -->|Inference| Vulkan[Vulkan Accelerated Inference]
    SystemArchitecture -->|Communication| PubSub[libP2P Pub/Sub Channels]
    SystemArchitecture -->|Retraining| FL[Federated Learning Nodes]
    RocksDB -->|CRDT Syncing| IPFS[IPFS Cache]
    FL -->|Model Updates| EdgeNode1[Edge Node 1]
    FL -->|Model Updates| EdgeNode2[Edge Node 2]