Loops HouseSafe Near

NEAR HACKATHON
Links
Team
1 member- RAOwner
Raphael Hoang
Overview
Safe Near is a security layer for NEAR, built for people and for AI agents. It reads every NEAR block live and answers “is it safe?” before anyone signs.
Safety Check and transaction firewall: paste any account, token, contract, link or transaction to get safe / be careful / danger in plain words. Wallets call the same check as an API before signing. Agent Guard (MCP): before an agent pays, guard_transaction applies the owner’s policy (USD and NEAR limits counting tokens and NEAR Intents balances, allowed receivers) and answers allow, ask_human or block. It fails closed, and gives the agent a function-call key instead of full access. Watching the chain: every new contract is audited from its WASM, every new token is followed from its first liquidity, and hacks are caught within seconds and traced across chains through NEAR Intents/1Click (e.g. the Binance hot-wallet drain and its laundering memecoins). An ontology layer gives one answer to “what is this account?”.
Safe Near Academy (B2B MCP): our audits and launch checks became 2,500+ graded tasks for training other AI agents. Contract security has 5 levels, memecoins 4. Answers are scored by fixed rules against code, chain and real price outcomes, with hidden exams, a fixed exam and a leaderboard beside our own gates.
The Academy runs as an autonomous business. A person sets the rules in one file; the agent sells certificates and passes (paid in NEAR, USDC or any chain via NEAR Intents), books its own NEAR AI cost, and pays out revenue daily through a Sputnik DAO. Every payout passes our own Agent Guard; the agent holds only a function-call key. Every sale and payout is on a public ledger.
Live: app.safe-near.org, Telegram @safenear_bot, API /api/v1, MCP /mcp and /b2b/mcp. Rules decide every verdict; NEAR AI explains and can only add risk.