⠿ swyrmcomputeBack to website
GETTING STARTED

Intelligence, served by gaming PCs.

SwyrmCompute is an OpenAI-compatible inference API. The model runs on PCs with AMD Radeon GPUs owned by people like you; we coordinate them, bill customers per token and pay hosts 80% of what their GPU earns.

The full system works end to end on real hardware, with real payments in USDT and USDC on Solana and BNB Chain. SWYRM is an SPL token on Solana with a fixed supply; it is not needed to use the API.

Two sides. One network.

  • Builders deposit USDT or USDC, call /v1/chat/completions and pay for the tokens our servers count.
  • Hosts install the agent on an AMD PC, serve requests when the PC is free, and receive automatic payouts in USDT or USDC.

How it works

  • One GPU, the whole model: each PC runs the complete model through Strata and serves one chat at a time.
  • Tiered memory: VRAM, RAM and NVMe work together, so a 12 GB GPU can run a large mixture-of-experts model.
  • Verification: hidden test requests, speed checks measured by our servers and a reference node keep hosts honest.
  • Idle time: with the owner’s consent, an idle GPU mines the coin the network selects, only when it beats the electricity cost.