Workflows overview

Chat is great for figuring something out once. A workflow is for when you need that something to happen every time — reliably, in the same order, without you driving it by hand. It’s a visual graph: boxes (nodes) that each do one job — call a model, run code, hit an API, send an email — connected so the output of one flows into the next.

Why use a workflow instead of chat

Repeatable

The same steps run the same way every time — no re-prompting, no drift.

Automatic

Run it on a schedule and have the result emailed to you, hands-off.

Composable

Mix models, code, your tools, and your data in one pipeline — and reuse one workflow inside another.

Auditable

Every run is saved with its inputs and per-step outputs, so you can see exactly what happened.

What a workflow is made of

A workflow is a set of nodes wired together. Each node has a type that decides what it does — take input, call a model, run Python, render a template, call a tool, branch on a condition, loop over a list, or call another workflow. You connect them by referencing one node’s output in another’s input, and Catalyst figures out the order to run them in.

A workflow can also take a picture as input and hand it to a vision model — see Give a workflow an image to look at.

The full list, with what each one is for, is in the Node reference.

Two ways to build one

• From the editor — drag nodes onto the canvas, wire them, and fill in each one. Start with Build your first workflow.
• From a chat — describe what you want in conversation and Catalyst designs it for you. See Create a workflow by chatting.

Either way you get the same thing: a saved workflow you can run, schedule, and edit.

Running and automating

Run a workflow on demand and watch each node light up as it executes, or set it on a schedule so it runs by itself. Workflows can email you the result and attach files, so a scheduled workflow becomes a report that just shows up in your inbox. And whether you start a run by hand or it fires on a schedule, a floating runs indicator keeps it visible from any page and notifies you the moment it finishes.

Tip

You don’t have to choose between models inside a workflow either. Point different model nodes at different providers and compare them in a single run — the screenshot above does exactly that, fanning one research step out to GPT, Sonnet, and Gemini.