G

Node Operator Guide

Node Operator Guide

Node Types#

Compute Nodes#

Provide CPU and/or GPU execution capacity for AI agent workloads (Layer 8), SDK compute tasks (Layer 3), and general protocol execution. Compute nodes execute agent workloads inside the GAR v1 sandbox — the sandbox is enforced by the protocol. Compute nodes cannot access workload content or private keys. Compensation basis: per verified execution unit (CPU-seconds or GPU-seconds), measured by the validator network. Minimum hardware (MVP defaults — subject to governance)

Spec

Minimum

Recommended

CPU

8-core x86-64

16-core+

RAM

16 GB

64 GB

GPU

Optional (CPU-only tier)

NVIDIA RTX 3080+ or equiv

Storage

100 GB SSD

500 GB NVMe

Network

100 Mbps symmetric

1 Gbps

Storage Nodes Provide data persistence for domain resolver records, Merkle snapshots, IPFS content pinning, and encrypted Workspace data sync. Storage nodes must implement availability proofs — cryptographic evidence that data is being stored and accessible. Compensation basis: per GB per period, weighted by availability proof score. Minimum hardware (MVP defaults — subject to governance)

Spec

Minimum

Recommended

Storage

1 TB HDD

10 TB+ NVMe

CPU

4-core

8-core+

RAM

8 GB

32 GB

Network

100 Mbps symmetric

1 Gbps

Uptime SLA

99.5%

99.9%

Bandwidth Relay Nodes Provide P2P relay and routing capacity for Gao Network (Layer 6), Gao Browser (Layer 2), and Meshii messaging relay. Compensation basis: per GB of verified data relayed, measured by delivery receipts. Minimum hardware (MVP defaults — subject to governance)

Spec

Minimum

Recommended

Network

500 Mbps symmetric

10 Gbps

CPU

4-core

8-core+

RAM

8 GB

16 GB

Public IP

Required

Static preferred

Uptime SLA

99.0%

99.9%

Hybrid Nodes Operators may run multiple node types on the same hardware if it meets the requirements for each type. Compensation is calculated independently per service type delivered.

Registration Flow#

# 1. Install Gao Node software
npm install -g @gao/node-cli

# 2. Initialize node
gao-node init --type compute|storage|bandwidth|hybrid

# 3. Generate node identity (separate from operator wallet)
gao-node keygen

# 4. Link to operator domain (.gao domain required)
gao-node link --domain operator.gao

# 5. Set pricing (USDC, market-determined)
gao-node config --price-per-unit 0.001 --token USDC

# 6. Register on-chain
gao-node register

# 7. Start node
gao-node start
# → Validator audits begin within 24 hours

A .gao domain is required for registration. Node keypair is separate from operator wallet — the node signs receipts with its own key; operator wallet handles payment settlement.

Service Level Agreement (SLA)#

Operators declare an SLA at registration. The validator network measures actual performance against declared SLA.

Metric

Compute

Storage

Bandwidth

Uptime target

99.0%

99.5%

99.0%

Response latency

< 500ms job acceptance

< 200ms proof response

< 100ms relay start

Availability proof cadence

Per execution receipt

Every 6 hours (MVP default — subject to governance)

Per delivery receipt

SLA performance affects reputation score, governance weight accumulation rate, position in protocol routing preferences, and eligibility for priority workload assignment. Operators who fall below declared SLA consistently receive reduced workload routing until performance improves.

Compensation Model#

Direct Usage Fees (Rail B)#

Consumers pay operators directly for verified service delivery. Payment flows via Layer 4 (Gao Payment) — non-custodial, on-chain, no intermediary.

Consumer requests service
  ↓
Protocol routes to best available node (reputation + price)
  ↓
Node executes workload
  ↓
Execution receipt generated (signed by node + verified by validators)
  ↓
Consumer pays node directly — USDC, per verified unit
  ↓
Receipt recorded — compensation complete

Rail A Bootstrap Compensation#

During early network formation, nodes with verified performance above the minimum threshold may receive Rail A protocol compensation. See proof-of-contribution.md for the full scoring and distribution model.

Pricing#

Operators independently set pricing. The protocol does not set minimum or maximum prices.

Node type

Unit

Compute (CPU)

per CPU-second

Compute (GPU)

per GPU-second

Storage

per GB per day

Bandwidth

per GB relayed

Prices are expressed in USDC. Operators may update pricing at any time with a signed update. Price changes take effect for new workloads; in-flight workloads complete at the quoted price.

Reputation System#

reputation_score =
  (uptime_score             × 0.35) +
  (latency_score            × 0.25) +
  (availability_proof_score × 0.20) +
  (fraud_penalty_deduction  × -0.20)

Reputation affects workload routing priority. High-reputation nodes receive more workloads at higher prices. Reputation is non-transferable — it is earned by the specific node identity only.

Exit & Offboarding#

gao-node deregister --domain operator.gao

Graceful exit:

  • In-flight workloads complete or hand off

  • No penalty for graceful exit

  • Node removed from routing after 24-hour cooldown (MVP default — subject to governance)

  • Reputation score preserved for potential re-entry Abrupt exit (node goes offline without deregistering) results in reputation penalty proportional to outstanding commitments.

Risk Considerations#

Node operation involves ongoing costs and market risks including: demand variability, competitive pricing pressure, infrastructure maintenance costs, technical failures, and regulatory developments. No compensation level is guaranteed. See proof-of-contribution.md for Rail A details and validator-spec.md for how service delivery is verified.