Cooperative Multi-Device Neural Network Intelligence System Via Inference Partitioning in a C#/.NET Framework

This repository and research paper contains the official prototype source code and supporting interface applications for the NightFrame project, as detailed in the included paper "Cooperative Multi-Device Neural Network Intelligence System Via Inference Partitioning in a C#/.NET Framework".

The NightFrame ecosystem introduces a decentralized, edge-native micro-service architecture designed to partition, distribute, and execute large language model (LLM) inference across a mesh of heterogeneous, consumer-grade hardware. By leveraging a robust C#/.NET foundation, the system dynamically assesses node capabilities (such as CPU, RAM, and GPU constraints) to shard inference tasks utilizing pipeline parallelism, bypassing the need for centralized cloud infrastructure.

Beyond core compute partitioning, the system introduces a decentralized trust model and a double-entry internal ledger to incentivize cooperative computation, all coordinated via high-throughput gRPC streams and low-latency SignalR telemetry.

Repository Contents (REPO.ZIP): This archive provides the complete set of artifacts necessary to evaluate, replicate, and expand upon the system architecture described in the manuscript. It includes:

• Orchestrator Services (C#/.NET): The core micro-services governing the mesh network, including the OrchestratorService (central coordination), ShardCoordinator (model-sharding engine), DroneRegistry (dynamic node catalog), LedgerService (contribution-credit token economy utilizing LiteDB), and CellularCoordinator (RF telemetry aggregation).
• Protocol Buffer Definitions (.proto): The schema files defining the gRPC communication protocols used for high-throughput node-to-mothership data streaming and shard payload delivery.
• Orchestrator Web Admin Console (Next.js): The web-based administrative dashboard (OWAC) designed for monitoring real-time telemetry, mesh health, node roles, and prompt submission statuses.
• Mobile Console (React Native/Expo): A mobile-first interface (mOWAC) designed for on-the-go monitoring of edge devices and ledger statuses.
• Agent 3 Desktop Client (WinForms/WPF): The fully featured Windows desktop application that allows host machines to register as compute nodes, share local hardware resources, interact with neural-mind chats, and track ledger credit accrual.