ComputeMesh Architecture & Integration Guide
Comprehensive developer references, OpenAI API compatibility specifications, and hardware node deployment manuals.
1. OpenAI API Drop-in Quickstart
ComputeMesh implements full drop-in compatibility with the standard OpenAI SDK and REST specification. You only need to set your base_url and pass your ComputeMesh API key.
from openai import OpenAI
client = OpenAI(
base_url="https://computemesh.inetconnector.com/v1",
api_key="cm_live_your_api_key_here"
)
response = client.chat.completions.create(
model="qwen/qwen2.5-7b-instruct",
messages=[{"role": "user", "content": "How does ComputeMesh layer sharding work?"}],
stream=True
)
for chunk in response:
print(chunk.choices[0].delta.content or "", end="")⚡ Live In-Browser API Playground
Connected to Live GatewayClick "Run Inference" above to stream tokens in real-time directly through the decentralized mesh.2. Decentralized Mesh Architecture
ComputeMesh connects consumer GPUs, professional cloud accelerators, and multi-GPU mining rigs through a low-latency, peer-to-peer execution topology. The scheduler evaluates node proximity, memory bandwidth, and thermal stability to dynamically allocate neural network layers.
├── Node A (RTX 3080 16GB) : Layers 0–12 (Embeddings & Attention Blocks)
└── Node B (5x 8GB Rig 40GB) : Layers 13–32 (Deep Blocks & LM-Head) ➔ Streamed Tokens
3. Pipeline Layer Sharding & PCIe 1x Physics
Unlike training workloads that require high-bandwidth all-reduce operations, autoregressive inference at batch size 1 only transmits the activation tensor of a single token (e.g. 8.2 KB for a 32B model). Across a 500 MB/s PCIe 1x mining riser, this transfer takes only 0.016 milliseconds, introducing zero human-perceivable token latency.
4. Hardware Provider Node Setup
Anyone with a modern GPU (NVIDIA, AMD, or Intel) can run a ComputeMesh provider agent. Nodes authenticate using asymmetric Ed25519 cryptography without exposing private keys.
5. Mining Rig NodeOS Appliance
NodeOS is a headless appliance image based on Debian 13 that boots directly from a USB stick. It features native dual-stack driver detection for both AMD (Polaris, Vega, RDNA) and NVIDIA (Pascal, Turing, Ampere) cards with an embedded real-time dashboard on port 8080.