Trading Guard.
Trade with natural language, without losing control.
An AI trading interface designed to protect users from themselves.
01 Context
Natural language can lower the friction of placing a trade. That convenience can also make impulsive actions easier.
02 Idea
Explore a Hyperliquid trading interface where an AI agent interprets intent while hard rules bound execution.
03 Constraints
The initial concept uses a $100 maximum position and 5× maximum leverage. These are proposed product limits, not a claim that trading within them is safe.
04 Exploration
The key question is where interpretation ends and authorization begins. Ambiguous instructions should produce a clarification, not an order.
05 Architecture
Proposed flow: natural-language intent → structured order proposal → deterministic risk checks → explicit confirmation → execution. The model should not be able to edit or bypass the limits.
06 Decisions
Risk constraints belong in deterministic code outside the model. Separate generating an order proposal from granting authority to submit it.
07 AI Workflow
AI is part of the proposed interface. Development workflow details and adversarial test results will be recorded once the prototype exists.
08 Result
Concept stage. No live trading system, performance results or execution guarantees are claimed.
09 What I’d change
Start with a simulated order flow and test rejected requests before connecting any execution capability.