> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://api.logicdialog.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://api.logicdialog.ai/_mcp/server.

# Alerts and notifications

Use workflows to turn conversation problems and trends into actions such as Slack messages, support tickets or emails. Alert collection is managed by LogicDialog. You do not need to configure monitoring services, thresholds or webhook URLs.

## Choose what to respond to

Add a **LogicDialog event** trigger to a workflow and choose an event:

| Event                      | When it runs                                                                      |
| -------------------------- | --------------------------------------------------------------------------------- |
| Conversation error         | A conversation reports a technical failure, a failed action or an analysis error. |
| Monitoring alert raised    | A trend alert becomes active, such as a spike in action or webhook errors.        |
| Monitoring alert resolved  | An active trend alert clears automatically.                                       |
| Monitoring alert dismissed | An active trend alert is dismissed by an operator.                                |

Conversation error relates to one conversation. Monitoring alerts relate to activity across an assistant, rather than one conversation. An unsuccessful business outcome alone does not count as a technical conversation error. A failed action can produce a conversation error even if the conversation recovered and continued.

## Send an alert to Slack

1. Open **Workflows** in a workstream and create a workflow.
2. Add a **LogicDialog event** trigger and select **Monitoring alert raised**.
3. Optionally add a condition to choose which alerts to act on.
4. Connect a **Slack** integration node, select a connection and choose the channel.
5. Set the message, save the workflow and apply the workstream to its destination environment.

For example, use these fields in a message:

```text
{{data.alert.title}}
Assistant: {{data.agent_id}}
Action: {{data.alert.tool_name}}
```

For conversation errors, use `{{data.conversation_id}}` and `{{data.error.message}}` instead. You can build separate workflows for raised and resolved alerts, or connect a different action such as creating a ticket. Alert workflows are notifications; an HTTP response node is not required.

## Decide which alerts need action

Monitoring event data includes:

| Field                     | Meaning                                                        |
| ------------------------- | -------------------------------------------------------------- |
| `alert.id`                | Identifier shared by an alert's lifecycle notifications.       |
| `alert.title`             | Human-readable description of the alert.                       |
| `alert.action`            | `triggered`, `resolved` or `dismissed`.                        |
| `alert.monitor`           | Stable alert category for conditions.                          |
| `alert.tool_name`         | Affected action, when the alert relates to one.                |
| `alert.occurred_at`       | Notification time in Unix seconds.                             |
| `agent_id`                | Assistant associated with the alert.                           |
| `destination_environment` | Environment in which this workflow is running.                 |
| `source_scope`            | `agent` for assistant-wide trend alerts.                       |
| `source_environment`      | `null` when the alert cannot be attributed to one environment. |

For example, match `alert.monitor` to `webhook_error_spike` to send webhook-related alerts to your technical support channel. Other categories include `elevated_conversation_failure_rate`, `elevated_tool_failure_rate`, `guardrail_trigger_spike`, `call_initiation_error_spike` and `low_evaluation_success_score`.

## Understand workspace and environment scope

Alerts are routed only to the workspace that owns the affected assistant. An alert for another customer's assistant does not trigger your workflows.

An assistant-wide trend can combine activity from multiple environments. Such alerts reach subscribed workflows in each deployed environment of the owning workspace, with `source_environment: null`. The workflow decides whether to act. For example, filter on `destination_environment` if only your Live workflow should send Slack notifications. This selects which workflow sends the notification; it does not prove the problem happened in Live.

Only deployed workflows receive alerts. Saving or previewing a draft does not subscribe it to operational alerts.

## Delivery and retries

LogicDialog verifies incoming alerts and deduplicates successful delivery to each destination. Raised, resolved and dismissed notifications are handled separately, and a later occurrence can trigger another run.

A workflow that fails after sending a message can repeat that action on retry. Where needed, use `alert.id`, `alert.action` and `alert.occurred_at` together as an idempotency key in the receiving system. Previewing an integration action can send a real notification.