Skip to main content
This walks you through bringing a node online end to end. Make sure you meet the requirements first. The node software is open source: github.com/Singularity-Layer/sgl-network-node. Full command details are in the CLI reference.
Prefer to let an agent do it? Install the x402-compute skill and ask your agent to bring the node online — it runs this whole flow (stake → install → register → attest → serve):
npx skills add https://github.com/ivaavimusic/singularity-skill --skill x402-compute
# or: curl -fsSL https://api.x402layer.cc/skill/x402-compute/install | bash

​
1. Install the node CLI

Install sgl from the official release (see the repo README for per-platform instructions). Verify it’s available: 
sgl --help
You also need a local inference runtime (llama-server) and a GGUF model file to serve, e.g.: 
# example: a small instruct model
~/models/Llama-3.2-3B-Instruct-Q4_K_M.gguf

​
2. Log in & register (with your staked wallet)

sgl login opens a browser flow to approve the node with the wallet that holds your stake. This binds the node to that wallet. 
sgl login --tee-type apple_se --models llama-3.2-3b
  • --tee-type — the TEE on your machine (default apple_se).
  • --models — comma-separated models you’ll advertise.
Prefer no browser? sgl init --wallet <STAKED_WALLET> registers directly. login is recommended.

​
3. Attest your enclave

Produce and submit the hardware attestation. Only attestation-verified nodes receive jobs — and you must re-attest after any binary update. 
sgl attest

​
4. Start serving

For production, run as a background service so the node survives reboots, logout, crashes, and idle sleep: 
sgl service install \
  --model-path ~/models/Llama-3.2-3B-Instruct-Q4_K_M.gguf \
  --model-name llama-3.2-3b \
  --resource-percent 50
To test in the foreground first: 
sgl start --model-path ~/models/Llama-3.2-3B-Instruct-Q4_K_M.gguf --model-name llama-3.2-3b
Check it’s healthy: 
sgl status            # node + hardware + orchestrator
sgl service status    # background service state
Your node now registers, advertises its model(s), heartbeats, and starts receiving jobs.

​
5. Maintenance

Going down on purpose? Use maintenance mode so you stop receiving new jobs cleanly (in-flight jobs finish; no penalty): 
sgl off-grid          # before planned downtime
sgl on-grid           # resume
Honest downtime is never penalized — it just means no new jobs while you’re offline. Only proven tampering is slashable (see Staking to operate).

​
Confidentiality

Your node runs every workload inside the TEE, so you serve user prompts without being able to read them. Attestation enforces this — it’s the basis of the network’s confidential-compute guarantee.

​
Next