⚡ Network Architecture

Democratizing AI Compute

SpazeNode aggregates idle GPUs around the globe, creating a secure, ultra-low-latency, and highly scalable inference engine. Here is how the magic happens.

storefront
Stage 01Service Layer

Developer Console & API

The entry point for developers. Spin up serverless inference API endpoints, deploy models via our command-line tool, or use the Python/TS SDKs to integrate compute nodes directly into your codebase.

  • check
    OpenAI-compatible chat & completions endpoint
  • check
    Deploy custom Docker weights with one command
  • check
    Managed API key access & granular team roles
  • check
    Live request monitoring & token usage statistics
developer-terminal (~/spazenode)
$ # Install SpazeNode CLI
npm install -g @spazenode/cli
$ # Log in with your API Key
spazenode auth login --key=sp_live_9e28fa0d
✔ Success! Authenticated as organization 'Alpha Labs'
$ # Deploy a Llama-3-8B endpoint on 2x RTX 4090s
spazenode deploy --model llama-3-8b --gpu RTX4090 --count 2
➜ Spin-up requested... allocating nodes...
✔ Endpoint live: https://api.spazenode.com/v1/chat/completions
settings_suggest
Stage 02Virtualization Layer

Decentralized OS

The brains of the network. This proprietary orchestration layer aggregates thousands of individual, heterogeneous GPUs into a single virtual data center. It performs routing, security checks, and hardware attestation.

  • check
    Latency-aware smart routing to the nearest active GPU
  • check
    Dynamic batching and KV-caching for up to 4x throughput
  • check
    Continuous secure hardware attestation & reputation checks
  • check
    Auto-failover: requests transparently re-route if a host disconnects
orchestrator-logs (region: us-east-1)
[INFO] Inbound request received for endpoint /llama-3-8b
[DISPATCH] Performing zero-trust hardware attestation...
➜ Attestation token verified: Host #4902-RTX4090
[ROUTING] Evaluating nodes latency & queue depth...
➜ Host #4902-RTX4090 (us-east-4a) - Latency: 14ms (Selected)
➜ Host #3812-RTX4090 (us-east-1b) - Latency: 28ms (Backup)
[BATCH] Applying dynamic request queue collation...
➜ Queue depth: 3 reqs -> combined into token batch (vram efficiency)
[DISPATCH] Forwarding payload to Host #4902 sandboxed docker container
hub
Stage 03Supply Layer

Distributed GPU Network

The physical compute infrastructure. Hosts from all over the world (small data centers, mining facilities, private dev rigs) install our sandboxed agent. The agent securely executes inference queries inside a firewalled container.

  • check
    Secure firewalled Docker runtime for complete host isolation
  • check
    Automatic payouts in stablecoins based on processed tokens
  • check
    Support for H100s, A100s, L40s, RTX 4090s, and RTX 3090s
  • check
    Automatic hardware benchmark and rating upon registration
host-agent-daemon (~/spaze-agent)
$ spaze-agent-daemon --status
● Daemon: Active (Listening for workloads)
Hardware: NVIDIA RTX 4090 (24GB VRAM) | Drivers: 535.11
Connection: Fiber 1Gbps Symmetric (Checked)
[DAEMON] Inbound workload received! Spawn sandbox container...
➜ Model weights: Llama-3-8b (Cached in local storage)
➜ Compute active. GPU Load: 94% | VRAM: 18.2 GB | Temp: 68°C
[DAEMON] Workload complete. Tokens output: 1,420 | Processing time: 0.8s
[EARNINGS] Earned $0.00071 for computation. Lifetime: $148.20
code

For AI Developers

Access the scale of hyper-scalers without the premium cost. Spin up OpenAI-compatible API servers, fine-tune models, or lease dedicated nodes with a credit pool that scales with your team.

Deploy your first endpoint arrow_forward
dns

For Compute Providers

Put your idle GPU hardware to work. Register as a host, run our fully secure sandboxed docker agent, set your availability times and target pricing, and receive automated payouts in USDC.

Monetize your GPU cluster arrow_forward

Ready to experience decentralized compute?

Start deploying serverless endpoints in seconds, or monetize your GPU infrastructure on the global SpazeNode network.

Deploy Your First ModelMonetize Your GPU