Keep the primary address out of the interface.
Agents use a replaceable public alias while authorized records preserve the real account relationship.

Give every AI agent isolated access, deterministic limits, and proof for every transaction.
Built for agent execution. Governed by human intent.
VeilGuard separates what an agent needs to execute from what it never needs to see.
The primary address stays out of the agent interface.
A simulated receipt appears after approval.
Ready for simulated policy evaluation.
Privacy is useful only when permissions stay explicit and enforceable.
Agents use a replaceable public alias while authorized records preserve the real account relationship.

Each agent receives its own account, permissions, budget, and kill switch.
Asset rules, approved venues, spending caps, and human approvals execute before every trade.
Requests, policy results, approvals, and human-signed testnet outcomes are bound to one integrity-checked record.
Natural language proposes the action. Deterministic policy decides whether it can happen.
Assign a dedicated account to an AI agent.
Set assets, venues, limits, and approval rules.
Let the human wallet sign the exact policy-bound Base Sepolia request.
Review a human-only chain receipt or an explicit non-broadcast demo digest.
Clear boundaries
Sensitive addresses stay out of AI-facing interfaces while authenticated human records remain complete and verifiable.
Privacy controls do not make transactions anonymous or remove legal obligations.
Start with isolated access, explicit limits, and proof for every action.