Triggers

Triggers are Valohai's native automation feature. They launch specific, pre-configured workloads when given conditions are met, no custom code required.

Triggers can launch executions, pipelines, fetch repository updates, and more.

Why Use Triggers?

Secure by default: Built-in authentication with flexible options (HMAC, JWT, static tokens)

No infrastructure needed: No servers to maintain, no polling loops, no message queues

Rate limiting included: Protect against accidental runaway costs

Visual configuration: Set up complex automation through the Valohai UI

Audit trail: Every trigger launch is logged for debugging and compliance

Three Types of Triggers

Scheduled Triggers

Run workloads on a recurring calendar schedule or cron expression.

Use cases:

• Retrain models nightly with fresh data

• Generate weekly performance reports

• Run batch predictions every hour

• Archive old datasets on the first of each month

Learn about Scheduled Triggers →

---

Notification Triggers

Launch workloads automatically when Valohai events occur.

Use cases:

• Process new dataset versions as soon as they're uploaded

• Deploy models immediately after approval

• Trigger validation when training completes

Supported events:

• Dataset version created

• Model latest approved version changes

• Execution/pipeline status changes

• And more...

Learn about Notification Triggers →

---

Webhook Triggers

Receive HTTP requests from external services to launch workloads on demand.

Use cases:

• Trigger from Slack commands (/train-model)

• Auto-fetch and run when code is pushed to GitHub

• Process images when labeling is complete (V7, Label Studio)

• Respond to custom application events

• Integrate with CI/CD pipelines

Supports industry-standard authentication:

• Hash-based Message Authentication Code (HMAC)

• JSON Web Tokens (JWT)

• Static secret tokens

• Timestamp validation for replay attack protection

Learn about Webhook Triggers →

---

How Triggers Work

Every trigger has two components:

1. Conditions (When to Launch)

Conditions validate whether the trigger should fire. All conditions must pass for the trigger to launch.

Schedule conditions:

• Cron expressions for flexible scheduling

• Simple recurring options (daily, weekdays, monthly)

Webhook conditions:

• Authentication (verify the request is legitimate)

• Timestamp validation (prevent replay attacks)

• Rate limiting (prevent excessive launches)

• Payload filtering (only trigger for specific events)

Notification conditions:

• Payload filtering (only trigger for specific datasets/models)

2. Actions (What to Launch)

Actions define what happens when all conditions pass.

Available actions:

• Run an execution

• Run a pipeline

• Copy a pipeline

• Fetch repository updates

• Return custom webhook responses

You can chain multiple actions in a single trigger.

---

Common Patterns

Pattern: Scheduled Training Pipeline

Goal: Retrain your model every night at 2 AM UTC

Setup:

1. Create trigger with type "Scheduled"

2. Add condition: Cron schedule 0 2 * * *

3. Add action: Run pipeline "nightly-training"

4. Save trigger

Result: Pipeline launches automatically every night. No manual intervention needed.

---

Pattern: Auto-Process New Data

Goal: When a dataset version is created, automatically run preprocessing

Setup:

1. Create trigger with type "Notification"

2. Add action: Run pipeline "data-preprocessing"

3. Configure payload input to receive dataset version URI

4. In notifications settings, route "dataset version created" events to this trigger

5. Save trigger

Result: Upload data → pipeline launches automatically → processed data ready for training.

---

Pattern: Slack-Triggered Inference

Goal: Team members can trigger inference from Slack with /predict <image-url>

Setup:

1. Create Slack app with slash command

2. Create trigger with type "Webhook"

3. Add authentication condition (HMAC with Slack signing secret)

4. Add action: Run pipeline "inference"

5. Configure pipeline to receive image URL from Slack payload

6. Point Slack slash command URL to trigger webhook URL

Result: Type /predict https://... in Slack → inference runs → results posted back to Slack.

---

Trigger Management

Create triggers: Project Settings → Triggers → Create Trigger

Monitor trigger launches: Project Settings → Triggers → ... menu → View Logs

Debug failed triggers: Check trigger logs for condition failures and action errors

Disable temporarily: Edit trigger and uncheck "Enabled"

Triggers that repeatedly fail will auto-disable to prevent spam. Re-enable them after fixing the issue.

---

Triggers vs. REST API

Not sure whether to use triggers or the REST API? Here's a quick comparison:

Feature	Triggers	REST API
Setup complexity	Visual UI configuration	Write custom code
Authentication	Built-in options	You implement it
Rate limiting	Included	You implement it
Event source	Valohai + webhooks	Anywhere
Flexibility	Pre-configured actions	Full programmatic control
Use when...	Standard automation needs	Custom integrations

You can combine both! Use triggers for common patterns, and the REST API when you need custom logic that triggers don't support.

---

Next Steps

1. New to triggers? Start with Scheduled Triggers for the simplest automation

2. Need event-driven workflows? Explore Notification Triggers

3. Building integrations? Check out Webhook Triggers