Nodes & the palette

Every workflow is built from nodes you add from the palette. The palette is grouped by what you're trying to do, not by jargon — but each group maps to a category with its own colour on the canvas.

When it runs

The trigger that starts a workflow.

• On a schedule (trigger.cron) — run every N minutes/hours, or on a cron expression.
• Run a command (trigger.command) — start the workflow by hand (or from a local command).

Get & handle data

The deterministic core — plain, free, predictable steps.

• HTTP request (task.http) — call an API.
• Run a command (task.command) — run a local shell command or script.
• Transform (task.transform) — reshape data with an expression.
• Branch (task.branch) — take different paths based on a condition.
• For each / Loop (task.foreach, task.loop) — repeat steps over a list. This palette entry is a step, not the Loop object — that's a whole graph with its own file and its own brain.
• Dedupe cache (task.cache-unseen, task.cache-mark) — only act on items you haven't processed before. See Dedupe & caching.

AI agents

• Run an agent (agent.run) — the opt-in AI step. Returns a structured decision or draft. See AI opt-in per step.

Hand off

• Run a workflow (task.run-workflow) — call another workflow as a sub-step, so you can compose larger automations from smaller ones.

Reading the colours

| Category | Colour | Examples | | --- | --- | --- | | Trigger | Amber | trigger.cron, trigger.command | | Data / deterministic | Cyan | task.http, task.transform, task.branch | | AI | Blue | agent.run | | Output / hand-off | Muted | task.run-workflow |

For the full configuration of each node, see the node reference.