ChainReflex-OS: Hardware-Accelerated Multi-Agent Swarm Runtime

High-throughput, low-latency multi-agent execution operating system mapped directly to local AMD ROCm GPU compute acceleration.

---

Overview

Sequential agentic AI frameworks often bottleneck on CPU memory transfers and unoptimized GPU compute scheduling when running multiple concurrent reasoning swarms.

ChainReflex-OS is a specialized execution runtime that maps parallel multi-agent graph workflows directly to AMD ROCm-compatible hardware. By optimizing VRAM allocations, batching intermediate agent inference requests, and decoupling state transitions from UI streaming, ChainReflex-OS achieves high-throughput local agent execution without cloud latency.

---

Runtime Architecture

graph TD
    A[Client Request / Swarm Task] --> B[ChainReflex Dispatch Daemon - TypeScript]
    B --> C{Task Execution Scheduler}
    C -->|Worker Node 1| D[ROCm Compute Engine - PyTorch / HIP]
    C -->|Worker Node 2| E[ROCm Compute Engine - PyTorch / HIP]
    C -->|Worker Node 3| F[ROCm Compute Engine - PyTorch / HIP]
    D --> G[Unified VRAM Tensor Cache]
    E --> G
    F --> G
    G --> H[LangGraph State Synchronizer]
    H --> I[Event-Driven Client SSE Stream]

Loading

---

Core Components

1. Dispatch Daemon (src/, clients/): High-concurrency TypeScript daemon managing client websocket connections and task queues.
2. ROCm Compute Engine (src/engine/): Python-based acceleration layer interfacing directly with ROCm HIP runtimes and PyTorch tensor operations.
3. Hardware Supervisor (setup_and_start.sh, start_backend.bat): Automated verification script validating ROCm driver initialization (rocm-smi), VRAM availability, and containerized dependencies.

---

Repository Structure

.
├── clients/                  # TypeScript and Python client SDKs
├── config/                   # Swarm definitions, model weights config, and scheduler parameters
├── demo/                     # Demonstration scripts and benchmark runs
├── deploy/                   # Container definitions and deployment configurations
├── src/                      # Core agent runtime and ROCm execution engines
├── setup_and_start.sh        # Linux ROCm environment initialization script
├── start_backend.bat         # Windows development startup script
├── Dockerfile                # Production container specification
├── render.yaml               # Cloud deployment blueprint
└── requirements.txt          # Python runtime dependencies

---

Getting Started

Hardware Prerequisites

• AMD Radeon / Instinct GPU supporting ROCm (or CPU fallback mode)
• Linux with ROCm 6.x drivers installed (or Windows WSL2 ROCm bridge)

Setup & Launch

# Clone the repository
git clone https://github.com/HamzaKhanBUIC/ChainReflex-OS.git
cd ChainReflex-OS

# Run the automated hardware check and environment setup
bash setup_and_start.sh

---

License

MIT License - see LICENSE for details.