Neuro-Bridge

Neuro-Bridge is a high-performance IPC (Inter-Process Communication) Bridge designed to break hardware isolation in Linux Chroot/Proot environments on Android.

This project is not limited to a single specific function. Neuro-Bridge acts as a "Hardware Proxy", enabling processes within a Linux container (Ubuntu/Debian) to offload compute workloads to native Android Host drivers (Vulkan/Adreno/Hexagon DSP).

This opens up unlimited possibilities: from AI Inference, General Purpose GPU (GPGPU), image processing, to graphical rendering experiments, without being hindered by kernel driver incompatibilities (KGSL vs DRM).

---

Core Philosophy

"The Brain in the Box, The Muscle on the Metal."

The Chroot environment (The Box) is excellent for software development but is blind to hardware. The Android Host (The Metal) has full hardware access but is limited in software. Neuro-Bridge unites these two.

---

Architecture

Neuro-Bridge works by separating Request from Execution:

1. Server (The Host Node)

   • Runs native on Android (aarch64-linux-android).
   • Exposes access to:
     • Vulkan Compute (via ash / wgpu).
     • OpenCL (if available in vendor lib).
     • Neural Networks API (NNAPI).
   • Acts as an "Executor" that receives raw data/commands.

2. Client (The Chroot Node)

   • A library/CLI binary that runs in Linux Chroot (aarch64-unknown-linux-gnu).
   • Packages instructions and data, sends them over a socket, and waits for results.
   • Can be integrated into other Python, C++, or Rust scripts.

3. The Pipeline

   • Communication via Unix Domain Socket (/dev/socket/neuro_bridge.sock).
   • Designed for high throughput (low latency serialization).

---

Potential Use Cases

Due to its universal nature, Neuro-Bridge can be developed for:

• AI/ML Acceleration: Running ONNX/TFLite models using Adreno GPU (similar to NCNN/MNN but via bridge).
• GPGPU Tasks: Performing heavy mathematical calculations (matrix multiplication, crypto operations) on the GPU.
• Image Processing: Sending raw bitmaps for processing by Android's ISP/DSP.
• Video Transcoding: (Experimental) Accessing Android's hardware encoder/decoder.
• Custom Driver Implementation: Creating "Virtual Drivers" on the Chroot side that offload rendering instructions to the Host.

---

Getting Started

Prerequisites

• Rooted Android Device (Snapdragon series recommended for Adreno/Hexagon support).
• Rust Toolchain (with aarch64-linux-android and aarch64-unknown-linux-gnu targets).

Build

# Clone Repo
git clone https://github.com/EAISD/Neuro-Bridge.git

# Build Server (Host Side)
cargo build --release --bin neuro_server --target aarch64-linux-android

# Build Client (Chroot Side)
cargo build --release --bin neuro_client --target aarch64-unknown-linux-gnu

---

Protocol Overview

Neuro-Bridge uses a flexible binary protocol (bincode). The command structure can be extended as needed by installed modules.

// Example of flexible structure
pub enum BridgeCommand {
    // Basic Diagnostic
    Ping,
    GetHardwareInfo,
    
    // Generic Compute Payload
    ExecuteCompute { 
        module_id: String, // e.g., "ai_engine" or "math_core"
        payload: Vec<u8>   // Raw data
    },
    
    // Future Expansion
    AllocateMemory { size: usize },
    WriteBuffer { id: u32, data: Vec<u8> },
}

---

Disclaimer

Neuro-Bridge is experimental low-level software.

• Directly accessing hardware drivers can cause kernel panics or reboots if incorrect instructions (Malformed Instructions) are sent.
• Use with caution on production devices.

---

EAISD - Experiment Artificial Intelligent Software Development