Features
Tensora combines EVM compatibility, OP Stack security, and decentralized AI inference into a single Layer 2. Each feature serves a clear technical or economic purpose.
Feature Overview
EVM Compatibility: Deploy Solidity contracts without modification. Use Hardhat, Foundry, Ethers.js, and Web3.py.
OP Stack Infrastructure: Inherit Optimism's audited sequencer, batcher, proposer, and bridge contracts.
BNB Chain Settlement: Finalize state roots and handle disputes on BSC, reducing L1 costs vs Ethereum mainnet.
Calldata DA: Store transaction batches as BSC calldata for high availability and low cost.
TORA Gas Abstraction: Users pay gas in TORA via an ERC-4337 Paymaster; no BNB required.
AI Subnet Marketplace: Permissionless creation of isolated AI markets with custom models and scoring rules.
Yuma Consensus: Validators commit-reveal scores to prevent collusion and ensure quality rankings.
Decentralized Rewards: Emission splits (18% miners / 41% validators / 41% delegators) with halving schedule.
Standard Bridge: Lock WTORA on BSC, mint on Tensora L2; withdraw with 7-day challenge window.
Upgradeable Governance: UUPS proxies with timelock-controlled upgrades and on-chain voting.
Detailed Feature Breakdown
Gas Abstraction via ERC-4337
Users submit UserOperation bundles signed with their account's private key. The ToraPaymaster validates TORA balance, pays BNB to the sequencer, and deducts TORA from the user's account. This abstracts away the dual-token model (BNB internal, TORA user-facing).
Rationale: OP Stack nodes expect ETH/BNB for gas metering. Paymasters let us keep that internal invariant while offering TORA-only UX externally.
AI Subnet Isolation
Each subnet is an independent smart contract with:
A set of registered miners (model providers)
Validators who stake TORA and score outputs
Hyperparameters (task timeout, bond amount, slashing rate)
A share of global TORA emissions based on subnet weight
Subnets cannot interfere with each other. A compromised model in Subnet A does not affect Subnet B's rewards.
Example Subnets:
Linguista: Translation and summarization
Visiona: Image generation and classification
Predictia: Time-series forecasting for DeFi
Yuma Consensus Anti-Collusion
Validators commit a hash of their scores, then reveal the scores in a later block. The protocol aggregates revealed scores using a trust-weighted average. Validators who consistently diverge from the consensus lose trust weight and earn fewer rewards.
Why commit-reveal? Without it, later validators see earlier scores and can copy them, free-riding on others' validation work.
Last updated