Turn your idle CPU or GPU into AI fuel. hyGrid is a peer-to-peer network that distributes AI inference, training, and data jobs across contributor machines — and rewards them for every FLOPs contributed.
hyGrid routes compute jobs from the Hypatia AI coordinator to contributor machines worldwide. You run the node client, it does the work, you earn credits.
Install the hyGrid node client for your OS. It runs as a background service and uses only idle resources.
Set your CPU cap (default 50%), RAM limit (2 GB), and battery minimum. The client never pushes past your thresholds.
The client connects to the hyGrid coordinator over TLS WebSocket. Jobs are dispatched, executed in sandboxed WASM or JSON task format, and results returned.
Credits are earned per FLOPs contributed. GPU jobs earn 3× the CPU rate. High-reliability nodes earn a 50% bonus on top.
hyGrid is designed around three principles: your machine stays in control, jobs execute safely, and every contributor gets paid fairly.
Jobs run in WASM sandbox or JSON task mode only. Python pickle execution is permanently blocked. Your machine never runs untrusted native code.
Hard caps on CPU (50%), GPU (50%), and RAM (2 GB) by default. The client idles automatically when battery drops below 20% or thermal sensors spike.
All communication between nodes and the coordinator is encrypted via TLS. Job payloads are signed and verified before execution.
Every node has a trust score. Nodes that fail jobs, cheat on results, or time out are progressively banned: 24h → 7 days → permanent.
The coordinator profiles each node's real-world performance and routes jobs to the fastest eligible worker. Redundancy and checkpointing protect long-running jobs.
Any machine can join. Windows, Linux, macOS, ARM, Raspberry Pi. The more contributors, the more AI capacity — and the more credits shared.
Node clients are lightweight background services. Install once, configure your limits, and start contributing.
Installs as a Windows Service. Runs silently in the background. GPU support via DirectML.
Coming SoonInstalls as a systemd service. CUDA GPU support available. Ideal for always-on server nodes.
Coming SoonLaunchAgent-based background service. Contributes idle Apple Silicon cores when plugged in.
Coming SoonOptimized for ARM SBCs. Low power, always-on contribution. Perfect for RPi clusters.
Coming Soon💡 While client packaging is in progress, you can manually connect a Python node using the open protocol. Contributors who join early receive a 2× credit multiplier for their first 30 days.
Choose your platform and follow the steps. The client connects automatically once configured.
Credits are calculated per FLOPs contributed and awarded automatically. They accumulate in your account and can be redeemed for priority job access or future marketplace features.
| Resource Type | Base Rate | GPU Multiplier | Reliability Bonus | Notes |
|---|---|---|---|---|
| CPU Compute | 1× base | — | +50% at ≥95% success rate | 1 credit per ~1 trillion FLOPs |
| GPU Compute (CUDA/ROCm) | 3× base | 3.0× | +50% at ≥95% success rate | Tensor ops weighted higher |
| GPU Compute (Apple Silicon) | 2.5× base | 2.5× | +50% at ≥95% success rate | Unified memory advantage |
| Early Contributor Bonus | 2× all | applies on top | applies on top | First 30 days after joining |
Credit rates may adjust as the network grows. Existing balances are always honoured at the rate they were earned. Full credit formula →