How Daily Loss Limits Shape Safe AI Trading Agents
Daily loss limits are a core safeguard for AI trading agents. This guide explains why they matter, how owners configure signed policies, and how they interact
Produced with automation, then checked by deterministic quality rules and an independent source-grounded review before publication.
- 01Daily loss limits cap the maximum capital an AI agent can lose in a single day.
- 02Owner‑signed limits are enforced by the agent key and cannot be overridden by the agent itself.
- 03Exceeding a daily loss limit typically triggers an emergency stop that revokes the agent key.
- 04Loss limits do not guarantee that positions will be closed automatically; owner review is required.
- 05Setting limits involves balancing protection against the risk of premature shutdown during volatile markets.
Daily loss limits define the maximum amount of capital an AI trading agent may lose in a single calendar day. By enforcing this cap, owners protect their portfolio from runaway losses caused by model deviation, market shocks, or unexpected execution errors. The limit is applied through an owner‑signed policy attached to the agent key, and once reached the system revokes the key and halts further activity. Because the policy is stored on‑chain, it remains authoritative even if the agent software is updated or redeployed.
Why are daily loss limits a fundamental control?
Losses can accumulate quickly when an algorithm trades frequently or when market conditions change faster than the model can adapt. A daily loss limit provides a hard ceiling that prevents a single day’s adverse outcomes from eroding the entire allocated capital. It also creates a clear, auditable boundary that can be reviewed after the fact, supporting compliance and post‑mortem analysis. Owners can compare the realized loss against the signed threshold to verify that the system behaved as intended.
How are daily loss limits implemented in practice?
Owners set the limit when they create the agent key, using the documented policy fields. The limit is expressed in the account’s base currency and is checked against the cumulative realized loss for the current day. If the loss reaches the threshold, the system initiates an emergency stop that revokes the calling key. The stop does not automatically close open positions; those require separate owner action, ensuring that the owner retains final authority over any remaining exposure.
- The owner‑signed policy includes a daily‑loss field measured in account currency.
- The limit is evaluated after each trade settlement, not just at order submission.
- Reaching the limit revokes the agent key, preventing further order signing.
- Open positions remain open until the owner reviews and decides to close them.
- The limit resets at the start of the next calendar day based on UTC.
What are the trade‑offs when setting a daily loss limit?
A tighter limit reduces exposure but may cause the agent to stop during normal market volatility, limiting potential upside. A looser limit allows more trading freedom but increases the risk of large drawdowns. Owners must consider their risk tolerance, the agent’s historical volatility, and the liquidity of the markets it trades. In addition, operational latency and settlement delays can affect when the loss is recognized, so limits should include a modest buffer for timing uncertainty.
- 01Risk tolerance: Conservative owners may choose a low percentage of allocated capital.
- 02Historical performance: Agents with high variance may need a larger buffer.
- 03Liquidity: In thin markets, partial fills can delay loss realization, affecting the timing of the limit breach.
- 04Operational latency: Delays in order confirmation can cause the loss to be recognized later than expected.
When should an AI trading agent pause after repeated errors?
Repeated execution errors such as rejected orders, stale quotes, or mismatched signatures can indicate a deeper problem with the model or the integration. An emergency stop triggered by a daily loss breach is one signal, but owners may also configure error‑based pauses. For guidance on error handling, see When Should an AI Trading Agent Pause After Repeated Errors?.
How do daily loss limits interact with other risk controls?
Daily loss limits are one layer in a multi‑tiered risk framework. They work alongside order‑size caps, daily notional limits, and expiry dates. Because each control is enforced by the same owner‑signed policy, they collectively shape the agent’s behavior without requiring the agent to manage them directly. Understanding the hierarchy of controls helps avoid unintended conflicts and ensures that a breach of one limit does not unintentionally disable another.
A daily loss limit is a safety valve, not a guarantee of profit or loss avoidance.
Can the daily loss limit be adjusted mid‑day?
Owners can issue a new policy with an updated limit, but the change only takes effect after the new key is authorized and the previous key is revoked. This ensures that any adjustment is deliberate and auditable.
Where can I learn more about complementary risk limits?
See the article Essential Risk Limits Every AI Trading Agent Should Enforce for a broader view of controls that work together with daily loss caps.
Frequently asked questions
The emergency stop revokes the agent key, preventing new orders. Existing positions stay open until the owner reviews them and decides whether to close or adjust them.
Owners can issue a new policy with an updated limit, but the change only takes effect after the new key is authorized and the previous key is revoked.
It sums realized losses from settled trades for the current UTC calendar day, using the account’s base currency. Unsettled trades are not included until they settle.
It caps the amount of loss that can be incurred in a day, but it does not eliminate market risk. Large adverse moves can still occur within the allowed loss amount.
See the article [Understanding Daily Notional Limits for Trading Agents](/blog/understanding-daily-notional-limits-for-trading-agents) for complementary guidance.
Sources and verification
Product claims in this article were checked against these first-party references. Runtime status remains authoritative for current availability.
- Felix documentationfirst party
- Felix machine referencefirst party
Build with Felix now.
Felix infrastructure is live through MCP and the API. The full trading app launches September 17.
Partial fills are common in automated trading. This article explains how an AI agent can reliably detect them, reconcile positions, respect limits, and manage uncertainty without assuming perfect execution.
A money map guides an AI trading agent through capital handling, risk controls, market data validation, and position tracking. This article outlines each element and its practical considerations.