A-Identity Guard for NEAR

A trust check for one question: should an AI agent pay this NEAR account?

Brief it with a NEAR account id (for example merchant.testnet or name.near) and, optionally, the amount.

You get back

• a verdict: ALLOW, WARN or DENY, with a 0 to 1000 score;
• every reason code behind it (account age and activity, recent code or key changes, full-access keys, Circle USDC blocklist, lookalike names of known accounts, verdicts in A-Identity's on-chain registry and in other teams' public registries);
• the block height it read and the sha256 of the snapshot of public chain data it used;
• the snapshot itself and the exact command that recomputes the same verdict from it, offline.

How it is built. The verdict engine is deterministic and open source (github.com/getA-Identity/A-Identity, folder near/). The same engine feeds a spend vault contract on NEAR: a DENY in the registry is refused on chain, a WARN or a payee nobody has judged is held for the owner, and the agent's only key is a function-call key that cannot move funds any other way. Testnet demo and a mainnet deployment of the reproducible build are linked from the repository.

Honest limits. It reads public data only. On a cited mainnet benchmark the engine alone warned on 20% of known scam accounts and denied none, which is why the vault holds unknown payees instead of trusting a score. Read the full benchmark in near/scan/benchmark/RESULTS.md.