Protocol Design

The R3MES Protocol

A decentralized AI training network with verifiable computation, efficient bandwidth usage, and fair economic incentives.

R3MES combines Proof of Useful Work with advanced cryptographic verification to create a trustless, efficient, and economically sustainable AI training network.

Architecture

Four interconnected layers powering decentralized AI

Compute Layer

Distributed GPU network for gradient generation and model training

  • Miner Nodes
  • Gradient Generation
  • BitNet LoRA Compression

Verification Layer

Three-layer optimistic verification for trustless computation

  • Merkle Proofs
  • Trap Jobs
  • Iron Sandbox

Consensus Layer

Cosmos SDK-based blockchain for coordination and settlement

  • CometBFT Consensus
  • IBC Compatibility
  • On-chain Governance

Economic Layer

Fair reward distribution and sustainable tokenomics

  • Role-based Rewards
  • Treasury Buy-back
  • Inference Fees

Key Innovations

What makes R3MES unique

Proof of Useful Work

Miners contribute real AI training work instead of wasteful computations, creating tangible value for the network while earning rewards.

Three-Layer Verification

Optimistic verification with Merkle proofs, random trap jobs, and Iron Sandbox ensure training integrity without sacrificing performance.

BitNet LoRA

99.6% bandwidth reduction through BitNet quantization enables efficient federated learning with minimal network overhead.

On-chain Governance

Token holders participate in protocol decisions through proposals and voting, ensuring community-driven development.

R3MES Token

The R3MES token powers the entire ecosystem - from mining rewards and staking to governance and inference payments.

Mining Rewards40%
Ecosystem Fund25%
Team & Advisors15%
Community10%
Treasury10%
1B
Total Supply
100M
Initial Circulating
10 R3MES
Block Reward
4 Years
Halving Period