Requirements

​

Hardware & TEE

• A machine with a supported Trusted Execution Environment (TEE). The enclave is what lets you serve confidential workloads without being able to read them — it’s mandatory.
  • Examples: Apple Secure Enclave (apple_se), Intel TDX/SGX, AMD SEV-SNP, AWS Nitro.
• Enough compute for the model you’ll serve — CPU/RAM (and GPU/Metal where available). Smaller models (3B-class) run on modest hardware; larger models need more.
• Reliable power + network — uptime is how you get matched to jobs (see Earnings).

​

Stake

• $SGL stake bonded to your operator wallet, at least the minimum required to register a compute node (shown in the staking app). This is your skin in the game.
• Staking is non-custodial and only slashable for proven tampering — never for honest downtime. See Staking to operate and the Staking Engine docs.

​

Software

• The sgl node CLI — github.com/Singularity-Layer/sgl-network-node (open source). See the repo README for install per platform.
• A local inference runtime (llama-server) for serving models.
• A model file (GGUF) for the model you intend to serve.

​

Wallet

• A Solana wallet holding your stake, used to authorize the node (browser-approved sgl login).

​

Checklist