Loops Houseanchorium-near-Paypal

NEAR HACKATHON
Links
Team
1 member- ADOwner
Adesh Shukla
Overview
Don't trust the bot. Verify it.
Problem: AI trading agents ask you to trust that they follow their own risk rules. When one breaks a rule, nobody sees it, its track record is a screenshot, and its kill switch lives in the same software it should stop.
What we built: Anchorium, an autonomous trading agent where NEAR enforces the rules, holds the money and verifies the AI.
- Guard contract: before every trade the agent journals its decision on-chain. The contract re-checks the owner's policy (asset, $ per trade, daily entries, cooldown, slippage, open positions, daily loss cap), computes P&L itself and pauses itself at the loss cap. The agent can veto, never override.
- Vault contract: holds the funds and pays a NEAR Intents 1Click deposit only for an ALLOWED, unpaid decision. On testnet a compromised agent asked for $4.99: rejected, the vault moved nothing. An allowed $2 decision released exactly 2 test-USDC. A replay and a paused vault were refused. Every tx is linked.
- The agent holds only function-call keys: it can journal and request a release, never move funds, pause or change rules.
- AI risk co-pilot on NEAR AI Cloud inside a TEE. make verify-ai checks the Intel TDX + NVIDIA attestation and each review's signature offline, and its hash is journaled on-chain.
- Swaps settle through NEAR Intents 1Click into the user's own account. An optional PayPal sandbox on-ramp funds the budget; NEAR still decides.
Proof, not promises: keyless make demo with real 1Click quotes against our live testnet guard; make vault-demo on your own testnet accounts; 454 Python tests, guard 40+2 and vault 16+1 contract tests, CI green. Two paper trades with every fee shown: one loss, one win.
Honest limits: testnet only, no mainnet trade yet; signed AI reviews wait for NEAR AI credits (attestation verified).