Owner Authority vs Agent Authority: Understanding the Core Differences

Owner authority and agent authority serve distinct roles in a trading‑agent system. Owner authority holds the master key that controls the underlying funds and can define policy limits for any agent. Agent authority, by contrast, is granted by the owner and is restricted to the actions explicitly allowed, such as placing trades within set boundaries. Understanding these differences is essential for managing risk and ensuring that automated activity remains under proper oversight.

What does owner authority actually control?

The owner key is the sole source of custody for the account’s capital. It can sign any transaction, including withdrawals, policy updates, and emergency stops. Because the owner holds this ultimate power, they can impose limits on agents-order size, daily notional exposure, daily loss caps, and expiry dates-through signed policy documents. These limits are enforced by the system at runtime, but they do not eliminate operational or market risk; a trade that complies with limits can still lose the full allocated amount.

How is agent authority defined and limited?

An agent key is created by the owner and carries a narrower scope. It may be authorized to submit orders, query market data, or adjust non‑financial parameters, but it cannot move funds out of the account. Withdrawal requires a separate owner‑signed intent, ensuring that even a compromised agent cannot empty the wallet. The agent’s permissions are bound by the policy fields the owner signs, and any deviation results in a rejected request.

• ·Trade‑scoped agents can only place orders within the defined size and notional limits.
• ·Agents cannot initiate withdrawals; that path is reserved for owner‑signed intents.
• ·Policy fields include order size, daily notional, daily loss, expiry, and custom limits documented in the official guides.
• ·Agent activity is logged and can be paused via an emergency stop, which revokes the agent key but leaves existing positions untouched.

What happens when an emergency stop is triggered?

An emergency stop immediately revokes the calling agent key, preventing further actions from that agent. It does not automatically close open positions or cancel token allowances that were previously granted. Those actions require a separate review and explicit owner transaction. This design protects against accidental loss of control while acknowledging that existing market exposure may remain until the owner decides how to resolve it.

Why are durable mutation identity and reconciliation important?

Because a timeout does not prove an order failed, the system records a durable mutation identifier for each request. This identifier allows the owner to reconcile the intended action with the actual outcome, ensuring that no trade is silently lost or duplicated. Reconciliation is especially critical when agents operate at high frequency, as missed acknowledgments can lead to unintended exposure.

• ·Every order carries a unique mutation ID for later verification.
• ·Owners should regularly compare mutation IDs with executed trades to detect gaps.
• ·Missing or unverified market data must be flagged rather than treated as zero.
• ·Reconciliation helps maintain accurate account state despite network delays.

How can owners set effective limits without over‑constraining agents?

Owners should balance risk controls with the agent’s operational needs. Setting limits too low may cause frequent rejections, while overly generous limits expose more capital to potential loss. Reviewing historical trade performance and adjusting limits periodically can help maintain this balance. For guidance on designing limits, see the article on daily notional limits for trading agents.

"Limits are a safety net, not a guarantee of profit or loss avoidance."

Further reading: Understanding Daily Notional Limits for Trading Agents and Why Consecutive Agent Errors Should Trigger a Pause. Also see the guide on How to Give an AI Agent a Trading Account for implementation details.