Automations
An agentic automation wraps one of your Nexa agents in a trigger and a set of actions, then deploys it as a scheduled or on-demand workflow. Use it when you want an agent to run unattended — for example, query your consumption (gold) tables every morning, generate a report document, and post the summary to Slack — instead of being invoked manually.
Automations live on the Agentic Automation screen (/agentic-automation in the web app). Each row is one automation; the list header describes them as agents that “monitor events, run jobs, and take intelligent actions.”
How an automation is composed
Section titled “How an automation is composed”Every automation is three things bound together:
| Part | What it is | Where it comes from |
|---|---|---|
| Agent | The reasoning engine that runs when the automation fires | An existing agent you built on the Agents screen |
| Trigger | The condition that starts a run | Chosen when you create the automation (see below) |
| Actions | Side effects the agent performs at the end of a run | Toggled in the automation config (documents, email, Slack) |
The agent supplies the intelligence; the automation supplies the when and the what happens next.
Trigger types
Section titled “Trigger types”You pick exactly one trigger type per automation:
| Trigger | Value | Fires when |
|---|---|---|
| On Demand | on_demand |
You (or an API call) request a run — no schedule |
| Scheduler (Time-Based) | cron |
A cron schedule elapses, e.g. every weekday at 8am |
| Custom Business Trigger | custom_business_trigger |
A SQL query you define returns rows on a polling schedule |
A Custom Business Trigger couples a SQL query with a cron polling interval: the automation runs the query on each poll and starts the agent when the condition is met. This lets you react to data — a threshold breach, a new record, a failed load — rather than only to the clock.
Actions
Section titled “Actions”After the agent finishes reasoning, the automation can perform one or more actions. The actions surfaced in the create wizard are:
- Generate Document — render a PDF from a document template (with an optional document type).
- Send Email Notification — email one or more recipients a subject and body.
- Slack channel notification — post to a configured Slack channel, optionally mentioning specific members.
Slack and email delivery rely on the Slack integration being configured. Document generation draws on your published document templates.
Status and lifecycle
Section titled “Status and lifecycle”The Status column reflects where the automation is in its build-and-deploy cycle, not whether it is currently executing:
| Status | Meaning |
|---|---|
| Draft | Created but not yet planned or validated |
| Not Deployed | Plan validated and/or artifacts generated; ready to deploy |
| Deployment Running | The GitHub Actions deployment workflow is in progress |
| Deployed | Live in the current environment; can run and be scheduled |
| Needs Attention | The last validation or deployment failed |
Once Deployed, an automation can be run immediately (Run Now), paused/resumed if it is scheduled, and promoted to the next environment. See Creating an automation for the full build flow and Promotion for moving an automation through dev, test, and prod.
Where automations run
Section titled “Where automations run”A deployed automation is materialized as a job on your data platform (Databricks or Snowflake), created and updated through a GitHub Actions workflow. The Job ID column links to the underlying job, and each automation’s detail page shows recent runs pulled from platform observability. This means automation runs are governed and billed the same way as your other jobs and pipelines.