How It Works — Agent Architecture | Hood Agent (HOOD4663)
// HOW IT WORKS

From message
to verified answer

One pipeline handles everything: your message goes to the agent, the agent calls tools, tools hit verified data sources, and anything touching money passes through the policy engine — in code and on-chain.

TELEGRAM DM or group chat AGENT (LLM) Claude or local Ollama proposes — never executes 14+ TOOLS every result: source + timestamp DATA SOURCES Chainlink · registry · GoPlus GeckoTerminal · Blockscout DefiLlama · RPC (viem) cached · rate-limited · circuit-broken POLICY ENGINE tier 1 auto · tier 2 confirm · tier 3 block SIMULATION every swap, before proposal BLOCKED honeypot · group chat · over budget ON-CHAIN EXECUTION session key · EIP-7702 · ZeroDev
// STEP BY STEP
{{ s.n }}
{{ s.t }}

{{ s.b }}

// THE STACK
{{ t }}
READ THE SOURCE ON GITHUB ↗ SECURITY MODEL →