Skip to content

Automation and AI

How Automation Runs

The short version — define, deploy, trigger, run, observe.

If you only read one page in this section, read this one. Everything else goes into detail on one of these five steps.

Define → Deploy → Trigger → Run → Observe

You build a Workflow Definition — a visual graph of steps (nodes) and connections (edges) in GreenVee’s workflow editor. This is a draft: editing it never affects anything currently running.

When you’re ready, you click Deploy. GreenVee validates the graph and, if it passes, freezes a numbered snapshot of it — a Deployment. This is the version that will actually execute; your definition stays freely editable, and future deploys create new, separate versions.

Every deployed workflow needs exactly one Trigger — Manual, Schedule, Webhook, or a data change on a table. The trigger is what decides when a deployment gets used to start a run.

When the trigger fires, GreenVee starts a run of the deployed workflow, following its steps in order. A step can call an AI model, read or write a record, branch, loop, wait, call an external tool, or hand part of the work to a subagent.

Every run — however it started, including chats with Polaris or an Agent — shows up live on Workflow Executions. You can watch its status change, drill into nested subagent runs, see a per-step trace, and export the result.

This five-step loop is about Workflow Definitions specifically. Agents are a different, complementary way to automate work — reactive rather than deterministic — and a workflow step can call one directly (a “Run agent” node). Agent Skills and MCP Servers are shared building blocks: the same Skill or Tool can be used by an Agent, by Polaris, or loaded mid-run by a workflow step.