Overview
The Logging & Alerts section provides centralized management for platform logging, operational alerts, and email notifications. Use this section to:- Send request and execution data to external logging and observability platforms
- Receive operational alerts through Slack
- Configure alert thresholds and reporting intervals
- Receive email notifications for selected platform events
- Logging Callbacks
- Alerting Types
- Alerting Settings
- Email Alerts
Logging Callbacks
The Logging Callbacks tab allows administrators to send platform activity and telemetry to supported external logging and observability systems.Add a Logging Callback
Select Add Callback to configure a new integration.- Select the required Callback from the dropdown.
- Enter the authentication and configuration information required by the selected integration.
- Save the callback.
- Verify that the callback appears under Active Logging Callbacks.
Credentials and connection information should be managed securely and limited to the permissions required by the integration.
Alerting Types
The Alerting Types tab controls which operational conditions generate Slack alerts.Slack Webhook Requirement
Operational alerts currently support Slack Incoming Webhook URLs. Before enabling an alert, create an incoming webhook for the Slack workspace and channel that should receive notifications. See the Slack Incoming Webhooks documentation for setup instructions. Each alert type can have its own Slack Webhook URL, allowing different operational events to be routed to different Slack channels.LLM Exceptions
Enable LLM Exceptions to receive notifications when model requests encounter exceptions. This alert is useful for detecting:- Model invocation failures
- Provider errors
- Unexpected inference errors
- Production request failures
LLM Responses Too Slow
Enable LLM Responses Too Slow to receive notifications when model responses exceed the configured performance expectations. Use this alert to monitor:- Provider latency
- Model performance degradation
- Unexpected response-time increases
LLM Requests Hanging
Enable LLM Requests Hanging to detect requests that remain unresolved for an unexpectedly long period. This alert can help identify:- Provider connectivity problems
- Long-running requests
- Timeout-related conditions
- Stalled model calls
Budget Alerts
Enable Budget Alerts (API Keys, Users) to receive notifications related to configured budget conditions. Use this alert with LLMGrid budget controls to monitor usage across users and API keys.Database Exceptions
Enable Database Exceptions (Read/Write) to receive alerts when database read or write operations encounter errors. This can provide early visibility into backend persistence or connectivity issues.Weekly/Monthly Spend Reports
Enable Weekly/Monthly Spend Reports to send recurring usage and spend reports through the configured Slack destination. This is useful for:- Platform administrators
- FinOps teams
- AI platform owners
- Usage governance reviews
Outage Alerts
Enable Outage Alerts to receive notifications when error activity indicates a potential model outage. Outage sensitivity is controlled through the settings available under Alerting Settings.Region Outage Alerts
Enable Region Outage Alerts to monitor provider-region availability. This can help identify incidents affecting multiple models within the same provider region.Configure Slack Alerts
For each alert type:- Enable the alert using its toggle.
- Enter the required Slack Webhook URL.
- Repeat for any additional alerts you want to enable.
- Select Save Changes.
Test Alerts
Select Test Alerts after configuration to verify that alerts can successfully reach the configured Slack destination.Testing alerts before relying on them for production monitoring is strongly recommended.
Alerting Settings
The Alerting Settings tab controls global operational alert behavior, reporting intervals, and outage detection thresholds.slack_alerting
Enables or disables Slack alerting for operational monitoring.
Slack alerts can cover conditions such as:
- Model outages
- Budget events
- Spend tracking failures
daily_report_frequency
Controls how frequently deployment latency and failure reports are generated.
The value is specified in seconds.
report_check_interval
Controls how frequently the background process checks whether a report should be sent.
The value is specified in seconds.
A shorter interval checks for pending reports more frequently.
budget_alert_ttl
Defines how long a budget alert remains cached.
The value is specified in seconds.
This helps prevent the same budget condition from generating repeated notifications within a short period.
outage_alert_ttl
Defines the error observation window used for model outage alerts.
The value is specified in seconds.
Errors occurring within this window contribute to outage detection.
region_outage_alert_ttl
Defines the error observation window used for provider-region outage alerts.
The value is specified in seconds.
Region outage alerts can identify failures affecting multiple models operating in the same region.
minor_outage_alert_threshold
Defines the number of qualifying errors required for a condition to be treated as a minor outage.
The UI notes that 400 errors are not counted toward this threshold.
major_outage_alert_threshold
Defines the number of qualifying errors required for a condition to be treated as a major outage.
The UI notes that 400 errors are not counted toward this threshold.
max_outage_alert_list_size
Controls the maximum number of errors retained for a model or region during outage evaluation.
This setting limits the size of the cached error list used for outage detection.
Update Alerting Settings
After modifying values, select Update Settings to apply the changes. Settings marked In Config indicate values currently defined through platform configuration.Email Alerts
The Email Alerts tab provides email notifications for selected administrative events. Email alerting has two parts:- Email Notifications
- Email Server Settings
Email Notifications
Select the events that should trigger an email notification. The UI currently provides notifications for:key_created
Sends an email notification when a new Virtual Key is created.
key_rotated
Sends an email notification when a Virtual Key is rotated.
invitation_created
Sends an email notification when a new user invitation is created.
This is particularly useful for tenant administrators managing user onboarding.
Save Notification Preferences
Select Save Changes after enabling or disabling notification events. Select Reset to Defaults to restore the default email notification configuration.Email Server Settings
Email notifications require an SMTP server. Configure the following fields:SMTP_HOST
Hostname of the SMTP server.
Example:
smtp.example.com
SMTP_PORT
Port used by the SMTP server.
The appropriate port depends on the SMTP service and its security configuration.
SMTP_USERNAME
Username used to authenticate with the SMTP server.
SMTP_PASSWORD
Password or credential used for SMTP authentication.
Treat the SMTP password as a sensitive credential.