Real assets,
agents on rails.
One sentence becomes a portfolio of real-world assets. AI proposes, the chain enforces.
A story on one side.
A spec on the other.
Anyone can read the thesis. The machine reads the manifest. Neither version can drift from the other, because the numbers on the right are the ones the vault actually obeys.
Asia Real Estate Income
Rental cash flow from apartment stock in Singapore, Ho Chi Minh City and Osaka, held through tokenized property funds and topped up with short-duration private credit.
Built for someone who wants a monthly-income shape rather than a growth curve, and who would rather give up upside than sit through a 20% drawdown.
"targets": { "weightsBps": [7000, 2000, 1000] },
"bounds": {
"floorBps": [6000, 1000, 500],
"ceilBps" : [8000, 3000, 2500]
},
"rebalance": { "trigger": "band", "bandBps": 300 },
"risk": { "maxDrawdownBps": 1500, "leverage": false },
"policy": {
"maxTxValue": "20% TVL", "dailyCap": "50% TVL",
"cooldown": "1h", "universe": "whitelist only"
}
Four parties.
None of them trusts the others.
The agent is the only part that thinks, and it is the part with the least power. Splitting these four roles is what keeps a compromised model, or a stolen key, from becoming a loss.
You
Policy
Agent
Vault
If the model returns nonsense, a deterministic rebalancer takes the turn instead. Nothing waits on a language model to behave.
Don't take the dashboard's word for it.
The vault holds tokens, and token balances are public. Read them straight off the chain and check the split yourself — the number in our interface has to match the number in your terminal.
$ cast call $VAULT "allocationOf()" \
--rpc-url rpc.testnet.chain.robinhood.com
What the chain can and can't prove
It can prove the vault is holding exactly the tokens and weights it claims, that the agent that signed each trade is the one registered to the strategy, and that no plan skipped a policy check.
It cannot prove that a token's issuer really owns the building behind it. That comes from attestations and audits, so the universe is restricted to issuers who publish them — and on testnet every asset is a simulation running on real historical NAV data.
Testnet build. Assets are simulated and priced from real historical series; points and balances carry no monetary value.
Strategies fill, then close.
Each launch has a hard capacity written into the contract. Deposit while a strategy is in its early window and the lower fee follows your deposit for as long as you hold it.
Steady Treasury Yield
Short-duration government paper and nothing else. The boring one, on purpose.
Asia Real Estate Income
Rental cash flow from apartment stock in Singapore, HCMC and Osaka, plus short-duration credit.
AI Adaptive Yield
Wide bands, agent's discretion. It moves between classes as spreads move, inside a fixed risk envelope.
Every decision, in the open.
Each line is a plan an agent signed and the chain accepted. Open one and you get the reasoning it was holding at the time, hashed into the transaction so it can't be rewritten after the fact.
We moved to the chain, not the other way round.
Running our own chain was never the product, and it made nothing better for anyone using this. RWAgent now runs where the tokenized assets it manages actually exist, and reads their real price feeds instead of a series we replay ourselves.
The tokenized shares an agent holds are quoted on this chain by their own feeds. Nothing is bridged, mirrored, or marked from somewhere else.
Each agent is a registered on-chain identity with a reputation that gates how much capacity it can launch.
Trades arrive as EIP-712 typed data. Wrong signer, stale nonce or expired deadline, and the call reverts.
A tokenized share trades every hour of the week. The share behind it trades 32.5 of them. We publish what happens in the other 135.
Start with a sentence.
Sign in with an email. Test funds arrive automatically. You'll be watching an agent work in about forty seconds.