Models propose. They do not execute.
Groq produces a structured proposal that is treated as untrusted input and reviewed by the hostile-input boundary before policy evaluation.
- SERVER-SIDE MODEL KEY
- STRICT STRUCTURED OUTPUT
- EXECUTABLE = FALSE
02ENGINE
HORRIS / ENGINE
THE EXECUTION ENGINE.
INTELLIGENCE UNDER CONTROL.
The Horris engine separates proposal generation from execution authority. AI can structure a trade idea; Horris policy, live preflight and explicit authorization decide whether it may advance.
Deterministic rules are authoritative.
Risk profiles, leverage limits, stop requirements, allocation caps and allowed actions are checked independently of model confidence.
- RISK CAPS
- ALLOWLISTS
- HARD REJECTIONS
Compile the exact transaction first.
Horris resolves live venue state, builds the exact call path and simulates it before any wallet-signing stage can be reached.
- LIVE STATE
- EXACT CALLDATA
- SIMULATION GATE
Authority is narrow and replay-safe.
EIP-712 authorization binds calldata, context, nonce and expiry so a previously approved action cannot be silently repurposed.
- NONCE
- EXPIRY
- CALLDATA HASH
Execution is not the end of risk.
The monitor continuously reasons about positions, stop coverage, frozen orders and new-risk freezes after entry.
- PROTECTION
- MONITOR
- RECOVERY
MODELGROQ / GPT-OSS
AI EXECUTIONDISABLED
SIGNINGEXPLICIT ONLY
FAILURE MODEFAIL-CLOSED
// HORRIS / SYSTEM PRINCIPLE
The model is never the wallet.
Horris keeps intelligence, policy, authorization and execution as separate authority layers so no AI response can silently become an onchain action.
[ OPEN CONTROL TERMINAL ]