> ## Documentation Index
> Fetch the complete documentation index at: https://docs.usetrident.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Set Up Signals to Monitor Agent Health

> Signals are no-code threshold monitors that alert you when agent error rates, latency, LLM cost, or finding counts cross a threshold.

Red-team evaluations and cloud scans give you point-in-time security assessments, but they don't tell you what's happening to your agent right now in production. Signals fill that gap. A Signal is a no-code threshold monitor that watches a specific metric — error rate, response latency, LLM spend, or new finding count — and fires an alert the moment that metric crosses a boundary you define. You configure a Signal once and Trident watches it continuously, routing alerts to Slack, PagerDuty, or email so your team can act before a minor spike becomes a major incident.

## Available metrics

| Metric              | Description                                                                  | Typical use                                                                |
| ------------------- | ---------------------------------------------------------------------------- | -------------------------------------------------------------------------- |
| `error_rate`        | Percentage of agent runs that result in an error, over a rolling time window | Detect regressions after a deploy                                          |
| `latency_p95`       | 95th-percentile response latency in milliseconds                             | Surface degradation before users notice                                    |
| `trace_volume`      | Total number of traces recorded per time window                              | Catch unexpected traffic spikes or drops                                   |
| `llm_cost`          | Estimated LLM API spend per time window, in USD                              | Set a spend ceiling and alert before budgets are blown                     |
| `negative_feedback` | Rate of negative user feedback signals attached to traces                    | Catch quality regressions from user reports                                |
| `new_findings`      | Count of new findings opened (filterable by severity)                        | Get an immediate alert when the firewall or a scan opens something serious |

## Create a Signal

<Steps>
  <Step title="Open the Signals view">
    In the [Trident dashboard](https://app.usetrident.dev), navigate to **Signals** in the left sidebar and click **New Signal**.
  </Step>

  <Step title="Select the metric and scope">
    Choose the metric you want to monitor from the dropdown. Then set the scope: either **All agents** in the project or a **specific agent** by name. Scoping to a single agent is useful when different agents have very different baseline characteristics.
  </Step>

  <Step title="Set the threshold condition">
    Define the condition that should trigger the alert. For example:

    * `error_rate > 5%` over a `15-minute` window
    * `latency_p95 > 2000ms` over a `5-minute` window
    * `llm_cost > $10` over a `1-hour` window
    * `new_findings >= 1` with severity filter `Critical`

    Select the comparison operator, value, and evaluation window from the controls.
  </Step>

  <Step title="Configure the alert destination">
    Choose where Trident should send the alert when the Signal fires:

    * **Slack** — select a channel (requires the [Slack integration](/integrations/slack-pagerduty))
    * **PagerDuty** — creates an incident (requires the [PagerDuty integration](/integrations/slack-pagerduty))
    * **Email** — sends to one or more addresses on the project

    You can configure multiple destinations for a single Signal.
  </Step>

  <Step title="Save the Signal">
    Click **Save**. The Signal becomes active immediately and begins evaluating on the next metric interval. Its initial state is **Active**.
  </Step>
</Steps>

## Signal states

| State        | Meaning                                                                                      |
| ------------ | -------------------------------------------------------------------------------------------- |
| **Active**   | The Signal is running and the metric is within the defined threshold                         |
| **Firing**   | The threshold has been breached; an alert has been sent to configured destinations           |
| **Resolved** | The metric has returned below the threshold; a resolution notification is sent if configured |

Trident sends a resolution notification automatically when a Firing Signal returns to the Active state, so your team knows the issue has cleared without manually checking the dashboard.

## Mute a Signal

During planned deployments or maintenance windows, you can suppress a Signal temporarily without deleting it:

1. Open the Signal from the Signals list
2. Click **Mute** and select a duration: **1 hour**, **8 hours**, or **24 hours**

The Signal continues to evaluate the metric during the mute window but does not send alerts. It returns to normal alerting behavior automatically when the mute period expires.

<Tip>
  Create a `new_findings` Signal scoped to **Critical** severity as your first Signal. This gives you an immediate notification any time the firewall or a red-team run opens a Critical finding — the one category where faster response time directly reduces blast radius.
</Tip>