Metacentre is an asset manager agent for tokenized stocks. Write what the book should look like, in plain English. The agent turns it into target weights, watches the gap between target and actual, and hands back the orders that close it.
| Target AAPL | 53.8% |
| Target GOOGL | 33.8% |
| Target NVDA | 12.4% |
| Largest drift | 14.7% |
| Proposed | sell 1.38 AAPL · buy 1.16 NVDA · buy 0.21 GOOGL |
A sentence of plain English becomes a working instruction: which assets are in scope, how much risk is tolerable, and what proportions were actually asked for. Percentages said out loud are kept as figures rather than paraphrased away.
Mean-variance optimization computes weights for the stated risk level. Those weights are then blended with the proportions requested, so the instruction shapes the outcome instead of being overwritten by the mathematics.
Where the computed target differs from what was asked for, the agent states both numbers and the reason for the gap. Nothing is presented as a match when it is not one.
Balances are read from the chain and compared against target. Once the gap passes threshold, the agent returns the exact orders that close it, sized and ready to sign.
Order, target, drift, and rebalancing, executed on-chain.
LiveTrend state, momentum, and volatility per holding, with stop and target levels derived from each asset’s own range rather than fixed percentages.
In developmentDeposit once and let the agent stand the watch, so allocation and rebalancing run without a signature at every step.
In developmentCoverage expands as more stocks gain on-chain liquidity deep enough to trade cleanly.