Build your first automation

The Automations area lists every workflow in your workspace with its status, trigger, and recent runs. You can start from a blank canvas or from a prebuilt recipe — failed-post alerts and weekly planning reminders are classic first automations because they pay off immediately and can’t do any damage.

For this guide, we’ll build a triage-style flow: when a new inbox conversation is classified as a complaint, notify the team in Slack. The same mechanics apply to any workflow you build later.

Every automation starts with exactly one trigger — that constraint keeps flows predictable. You choose between workspace events (a post published, failed, or approved; a channel connected; an agent run finished; inbox events), a cron-style schedule, or a manual Run button for flows you want to fire on demand.

On the canvas, the trigger is the leftmost node; everything downstream receives its data. You reference that data anywhere with simple {{expressions}} — the post’s title in a Slack message, the conversation’s intent in a condition.

Click any node to open its editor and configure what it does. Actions include creating a draft post, running an AI agent, sending a Slack, Telegram, or in-app notification, and calling any external API with the HTTP node (secrets are stored encrypted and referenced safely). If-conditions branch the flow, and Wait nodes pause a branch for minutes or days when timing matters.

One rail worth knowing before you build anything ambitious: the create-post action always produces a draft that enters your normal approval workflow. By design, no automation can publish directly to your channels — details on the Automations page.

Before activating anything, run the workflow in test mode. Test runs execute the whole graph synchronously: read-only steps run for real, while anything that would write — a draft, a notification, an API call — is simulated and clearly labeled. Each node shows success or failure right on the canvas, so you can validate the logic and the branching without sending a single message or spending credits.

Read the test results node by node: check that the trigger data contains what you expected, that each condition took the branch you intended, and that the simulated writes — the labeled Slack message or draft — look right with real values interpolated into them. If a condition routes the wrong way or an expression comes back empty, fix it and test again. This loop is cheap on purpose: it costs nothing and sends nothing.

When the test run looks right, activate the workflow. Live runs execute in the background, and every execution is recorded with a per-node trace you can audit later; runs can be cancelled mid-flight if needed. Admin and editor roles gate who can build versus activate, which matters once several people share a workspace.

From here, the interesting flows combine pillars: trigger an agent run when an upload is scheduled, wire inbox triage into Slack alerts, or use comment-to-DM keyword automations on your posts.