Creating an Automation

This page walks through creating an agentic automation in the Nexa web app. You start from an existing agent, attach a trigger and actions, review the execution plan Nexa generates, and then commit and deploy the automation as a job. Authoring happens in the dev environment — the Create Automation button appears only there.

Note

Before you start, you need a working agent. Automations bind to an agent you have already built on the Agents screen — the agent supplies the reasoning; the automation supplies the trigger and post-run actions.

The build flow at a glance

Creating an automation is a three-stage wizard, driven by the tab strip at the top of the create screen:

1. Automation Setup — identity, trigger, and actions. Ends with Show Plan.
2. Review Plan — the generated execution plan. Ends with Generate Automation Artifacts.
3. Deployment — live tracking of the GitHub Actions deployment workflow.

You cannot skip ahead: the plan must be generated before you can review it, and artifacts must be generated before deployment tracking opens.

Step-by-step

1. Open the create wizard. On the Agentic Automation screen, click Create Automation (visible in dev). This opens Configure Agentic Automation at /agentic-automation/create.

2. Fill in basic information. Provide the core identity of the automation:

   Field	Required	Notes
   Automation Name	Yes	2–50 characters; letters, numbers, spaces, and hyphens only. Cannot be all numbers, and cannot start with a number or hyphen.
   Description / Purpose	Yes	What the automation does and why.
   Agent	Yes	Select from your existing agents.
   Automation Prompt	Optional	Extra instruction passed to the agent at run time.

   A URL-safe automation alias is derived from the name automatically.

3. Choose a trigger. Select one trigger type and configure it:

   Trigger	Configuration
   On Demand	No schedule — runs only when requested or via Run Now.
   Scheduler (Time-Based)	A cron schedule built with the cron generator, plus a timezone (default America/New_York).
   Custom Business Trigger	A SQL query plus a cron polling schedule. The automation runs the query on each poll and fires the agent when it returns rows.

   For a Custom Business Trigger, click Validate to run the SQL against your warehouse and confirm it executes before continuing. The plan step is blocked until the query validates.

4. Configure actions. Toggle what happens when the agent finishes:

   • Generate Document — choose a document template (and optionally a document type) to render a PDF.
   • Send Email Notification — enter comma-separated recipients, a subject, and a body. Enabling document generation also enables email so the document can be delivered.
   • Slack channel notification — pick a configured Slack channel and, optionally, members to mention.

   Slack channels are loaded from your Slack integration; the channel and members must resolve before the plan can be generated.

5. Show the plan. Click Show Plan. Nexa saves a draft, then generates an execution plan — the ordered set of tool calls the agent will make (query, generate document, send email, post to Slack, and so on). The plan is shown three ways: a Graph, a Steps breakdown, and raw JSON. The header reports the number of steps, whether execution is sequential or parallel, and a confidence percentage.

6. Review the plan. On the Review Plan tab, confirm the steps match your intent. Editing any setup field invalidates the plan — you must regenerate it before continuing. When the plan looks right, click Generate Automation Artifacts.

7. Generate artifacts and deploy. Generating artifacts validates the automation definition (its metadata.json and automation.yaml), commits the files to your repository, and dispatches the GitHub Actions workflow that creates the underlying job. The wizard advances to the Deployment tab and polls deployment progress.

What “generate” actually produces

Generating artifacts is a two-phase operation:

1. Validate — the automation definition files are checked. If validation fails, the errors are shown and nothing is committed.
2. Create — the validated files are committed to your repository and a GitHub Actions run is dispatched to materialize the job on your data platform.

The automation’s Validation & Artifacts tab (on the detail page) later shows the target branch, the commit SHA, and the list of generated files.

Databricks

The deployment workflow creates a Databricks job for the automation. Recent runs on the automation’s detail page are pulled from Databricks observability, showing the run’s life-cycle state, result state, and per-task detail.

Snowflake

The deployment workflow materializes the automation as a job against Snowflake. Usage from automation runs is billed as Snowflake credits and appears in Usage & Costs.

After deployment

Once the status reads Deployed, you can:

• Run Now — trigger an immediate run (available when the job exists).
• Pause / Resume — for scheduled automations, stop or restart the cron trigger without deleting the job. Run Now still works while paused.
• Promote — copy the automation to the next environment via a pull request. See Promotion.

Next steps

• Promote the automation from dev to test to prod.
• Automations overview — triggers, actions, and status model.
• Configure the Slack integration so Slack actions can deliver.