Live on Mantle Mainnet
ERC-8004·agent #1The AI treasury
that proves itself.
Mensa allocates your funds across mETH and USDY on Mantle, then competes against humans in a tournament to prove every decision. Every choice is explained, logged on-chain, and challenged.
Built forDAO treasuriesSophisticated DeFi saversRWA-backed protocols
$0.00
TVL
mETH + USDY in agent
—
AI Win Rate
0W / 0L of 0 rounds
—
Alpha vs 50/50
building track record
Live Tournament
AI vs Human, same inputs, same time horizon, settled on-chain.
Loading rounds...
What Mensa is thinking
The latest on-chain decision, with reasoning. No black box.
Agent reasoning...
What Mensa does
Six features. Each one is built around Mantle's unique stack.
Auto-allocates between yield and safety
The agent reads live APYs and ETH macro signals to rebalance the treasury between staked ETH and T-bills. Targets the best risk-adjusted return at every step.
mETH (Mantle native staking) · USDY (Ondo on Mantle)
Explains every decision in plain English
Each rebalance ships with a written rationale. No black box: confidence score, market snapshot, and reasoning are all stored next to the action.
DecisionLog contract · low-gas full transparency
Proves itself in a Turing tournament
Anyone can stake their own allocation against the AI’s on the same round. After settlement, the closer-to-optimal allocation wins. No subjective judging.
TournamentVault contract on Mantle
Sybil-resistant human voting
Vote weight scales with sqrt(reputation), so 100 fresh wallets can’t outweigh one reputable voter. Whales and bots both get diminishing returns.
Reputation contract · min-stake gate via MensaAgent
Pays humans who beat the AI
Mensa charges a 15% performance fee — only on yield, never on principal. The fee funds a bounty pool that pays out to humans who out-allocate the AI.
BountyPool contract · 50/30/20 winners/reputation/ops
Verifiable performance vs passive HODL
Every settled round computes the AI return and the 50/50 baseline return from on-chain prices. The cumulative alpha — and the AI win rate — are public stats anyone can audit, not numbers we report.
Alpha measured per round · annualized after 30 rounds