Aureus

The problem: an AI that manages money leaks your strategy twice. Your mandate, holdings and every thesis go to a model provider in plaintext, and every swap on a public chain links your wallet to your positions, so anyone can map your book and front-run your next move. That keeps serious capital away from AI agents. What I built: AUREUS, an autonomous AI portfolio manager that keeps your strategy private and stays inside limits you set. You fund a trading wallet, write a mandate and set hard limits. Five agents then run every cycle: Scout reads prices and private balances, Strategist proposes target weights, Risk can only lower risk, Guard enforces per-trade and daily caps, a cooldown and a drawdown kill switch in code, and Executor trades. You keep your own keys, and deleting the agent's key locks it out instantly. A private dashboard lets you watch every cycle, pause or resume the agent, and disclose any decision. It runs live on NEAR mainnet every 30 minutes. How NEAR fits in: all reasoning runs in TEEs on NEAR AI Cloud. For every call the agent verifies a fresh attestation and the enclave's signature, and refuses to act on unverified output. Execution uses NEAR Intents 1Click with Confidential Intents: the book lives in the private shard and each trade is a NEP-413 intent the agent signs, so balances, history and routes never appear on a block explorer. Every decision is hashed, and one Merkle root is anchored on NEAR mainnet, so any single trade can be proven against the chain without revealing the others. NEAR accounts and access keys give the agent a revocable key. Every layer is NEAR native, from inference to settlement to proof.