> ## Documentation Index > Fetch the complete documentation index at: https://dev-docs.altrum.ai/llms.txt > Use this file to discover all available pages before exploring further. # Alerts > Get told when a budget, a guardrail, your traffic or an AI system needs you, in the bell and by email, and choose the thresholds. Altrum checks a fixed set of rules every hour. When one holds, everyone it concerns gets a notification under the bell at the top of every page, and an email if they want one. Each alert names the thing, the fact and the threshold ("Claims bot used 80% of its \$4,000 monthly budget"), and opens the screen where it is fixed. Alerts are notices. The AI Gateway already enforces budgets, limits and guardrails on every request, so an alert can arrive up to an hour after the fact. ## Where alerts show * **The bell** shows how many are unread (up to 9+) and the latest 8 open ones. **Mark all read** and **View all** are at the top and bottom. * **Alerts**, under Monitor in the sidebar, lists your own alerts, newest first, under Today, Yesterday and Earlier. Filter by severity and by rule. Click a row to open where it is fixed; it is marked read. Owners and admins also see **Create incident** on an open alert, which opens an [incident](/incidents) filled in from it. An alert **clears** by itself when its condition stops holding, for example when a budget resets or is raised. Cleared alerts move to **Cleared** and stay there as history. Notifications are read or unread; there is nothing to dismiss. ## The rules | Rule | Fires when | Starts at | Severity | Email to start with | | - | - | - | - | - | | Budget used | A budget reaches a level of its amount: 50, 80 or 100% | 80% and 100% | Warning, High at 100% | On | | Budget heading over | Spend is heading past a budget by a share before it resets (1 to 100%) | 10% | Warning | On | | Unusual spend | Today's spend, for the organization or an AI system, is over a multiple of its 7 day daily average (2 to 20) and over an amount (\$1 to \$100,000) | 3 times and \$20 | Warning | On | | Violation spike | An AI system's guardrail violations in 24 hours are over a multiple of its usual day (2 to 20), and at least a number (5 to 10,000) | 3 times and 20 | Warning | On | | Watched guardrail | Any violation of the guardrails you pick, on prompts, replies or both, for AI systems rated at a level and above | Data protection, on replies, High and above | High | On | | Error rate | An AI system fails over a share (1 to 100%) of at least a number of requests (10 to 100,000) in an hour | 10% of 50 | Warning | On | | Provider failing | A model provider errors or times out on over a share (1 to 100%) of at least a number of requests (10 to 100,000) in an hour | 20% of 20 | High | On | | Model refused | Requests were refused in the last 24 hours because their model is not allowed | Any | Info | Off | | Posture drop | An AI system's posture falls by a number of points since the last day measured (1 to 100), or it needs approval again after its rating rose | 10 points | Warning | On | | Reassessment due | An assessment is due within a number of days (1 to 90), or overdue | 14 days | Info, Warning when overdue | On | | Waiting for approval | An AI system has waited in review for over a number of days (1 to 30) | 3 days | Info | Off | | Unused key | A key has not been used for a number of days (7 to 365) | 90 days | Info | Off | | Unregistered AI | A key with no AI system sends over a number of requests in a day (10 to 100,000), once for each key, never for traffic dismissed in [Discover](/discover) | 50 | Info | Off | A budget alert counts only the budgets the AI Gateway enforces: the organization's, an AI system's and a key's. A [project's](/projects) budget is a sum and is never alerted on. When a budget reaches 100%, the 80% alert for that period clears, so you see one. **Access review due** is not on the Rules tab either: in the last 7 days of a quarter with no completed [access review](/access-reviews), it goes to owners and admins of an organization with the AI Gateway, once a quarter, and clears when a review is completed. It cannot be switched off, and does not email unless you switch it on. **Incident assigned** is not on the Rules tab: it is sent when an [incident](/incidents) is assigned to you, never by the hourly check, and clears when the incident is resolved or assigned to someone else. It emails by default. An organization with only the AI Gateway has no Posture drop, Reassessment due or Waiting for approval rules. One with only governance has those three and none of the others. ### How often each fires Each alert fires once for its window, however many times the hour comes round: * A budget alert, once per level per budget period. * Unusual spend, a spike, a watched guardrail, the error rate, a failing provider and a refused model, once a day for each AI system, provider or model. * A reassessment, once when it is due soon and once more when it is overdue. * Waiting for approval, once each time an AI system goes into review. * An unused key, once until it is used again. ## Who gets each alert | Alert about | Who gets it | | - | - | | Anything | Every owner and admin | | An AI system | Also its owner and its project's lead | | A personal key | Also the person it belongs to | | An AI system's key | Also the AI system's owner and its project's lead | **Incident assigned** is the exception: it goes only to the person the incident is assigned to. **Access review due** goes to owners and admins only. Only active members are told. ## Change a rule Owners and admins change rules; everyone else sees them read only. The rules are grouped by Spend, Guardrails, Traffic, Governance and Keys, each with what it fires on. A rule switched off fires nothing, and its open alerts clear on the next check. Select the pencil, change the numbers within their bounds, and select **Save**. A value outside the bounds is refused, naming the field and its range. Each change is recorded in the [audit trail](/audit-trail) as "Changed an alert rule". ## Choose which alerts you get by email Open the account menu (your initials, top right) and select **Email settings**. Each rule has a switch, saved as you change it. The bell always gets every alert, whatever you choose here. The email itself is described in [Emails](/reference/emails). This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.