Skip to content

💬 Slack Integration

What you'll learn

How to send pipeline alerts to Slack channels — don't watch the terminal, let the bot tell you when it's done.

Get notified where you work. Receive alerts for pipeline successes, failures, and approval requests directly in Slack.

Why Slack Alerts?

Feature Benefit
Real-time Know immediately when a production job fails
Visibility Keep the whole team in the loop
Actionable Links to logs and dashboards in the message
Low friction No context switching — alerts come to you

💬 Setup

1. Create a Slack Webhook

  1. Go to Slack API: Incoming Webhooks
  2. Create a new app or use an existing one
  3. Enable Incoming Webhooks and generate a webhook URL
  4. Copy the URL (format: https://hooks.slack.com/services/T.../B.../...)

2. Configure FlowyML

1
2
3
4
5
from flowyml import configure_notifications

configure_notifications(
    slack_webhook="https://hooks.slack.com/services/T000/B000/XXXX",
)

Or via environment variable:

1
export FLOWYML_SLACK_WEBHOOK=https://hooks.slack.com/services/T000/B000/XXXX

🔔 Sending Alerts

From Steps

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
from flowyml import get_notifier, step

@step
def notify_team(metrics):
    get_notifier().notify(
        title="🚀 Training Finished",
        message=f"Accuracy: {metrics['acc']:.2%}\nF1: {metrics['f1']:.2f}",
        level="success",
        channels=["slack"],
    )

On Failure (Automatic)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
from flowyml import step, on_failure

@step(
    on_failure=on_failure(
        action="slack",
        recipients=["#ml-ops"],
        include_logs=True,
    )
)
def critical_step():
    """Automatically sends a Slack alert with logs if this step fails."""
    return train_model()

Alert Levels

Level Slack Color Use Case
"info" Blue Pipeline started, progress updates
"success" Green Training complete, deployment successful
"warning" Yellow Drift detected, degraded performance
"error" Red Step failure, pipeline crash

Best Practices

Channel strategy

Use #ml-ops for all alerts, #ml-alerts-critical for errors only. Avoid sending success messages to high-traffic channels.

Webhook security

Store webhook URLs in environment variables or secrets — never hardcode them in source code.