Flows and nodes
A flow is drawn on a canvas as a chain of nodes. The first node is the trigger, and every node after it is a step. AI Studio runs the steps in order, following the lines that connect them.
Adding steps
Click Add Step on the canvas to open the Add a step picker. Nodes are grouped into tabs:
| Tab | What's in it |
|---|---|
| AI | Language models and AI tools, such as LLM and OpenAI. |
| DECA Cloud | Other DECA products: CRM, Tables, Knowledge Base, Forms, Chatbot, LiveChat and more. |
| Built-in | General-purpose tools: HTTP requests, code, formatting, filtering, branching, loops, waits, variables and files. |
| Apps | Outside services, such as Slack, Gmail, Google Calendar, Salesforce and LINE. |
Pick a node, then pick one of its actions. A node groups everything you can do with one product or tool, and each action is one operation. The Tables node, for example, has actions such as Create Record, Update Record and List Records. A node with only one action, such as Stop, skips the second screen.
Every node and action is listed in the Node catalog.
Configuring a step
Select a step to open its settings panel. What you see depends on the action, but most steps have two parts:
- Credential: the saved connection the step uses to sign in to another service. Pick an existing one or create one from the panel. See Connect a credential.
- Action settings: the fields the action needs, such as a channel ID, an email subject or a URL. Fill in the required fields before you run the flow.
Most fields can mix typed text with data from earlier steps. Type $ in a field to insert a value from the trigger or a previous step. See Using data from earlier steps.
Some fields have a Fixed / Expression switch. Choose Fixed to pick a value from a list or type it in. Choose Expression to fill the field from earlier steps instead.
Give steps clear names with Rename step. The name appears on the canvas and in the data picker, so Find customer is easier to work with later than Tables 2.
Outputs
When a step runs it produces an output, the data it got back. A List Records step outputs the records it found, an LLM step outputs the text it generated, and an HTTP step outputs the response status, headers and body.
Any later step can use any part of that output. The Node catalog lists the output fields of every action.
When a step fails
By default a failed step stops the flow. You can change this per step in Nodes Setting under Error Handling:
| Option | What happens |
|---|---|
| Stop flow (recommended) | The execution stops at the failed step and is marked as failed. |
| Continue flow | The flow carries on with the next step. Later steps may get incomplete data, so check before relying on the failed step's output. |
The Executions page shows which step failed and why.
Testing while you build
You can run a flow from the editor to check it. The flow needs a Manual trigger for this.
If a step would do something you don't want during a test, such as sending a real email, give it a Sample Payload in Nodes Setting. When you click Run, the step returns that sample as its output instead of running. Webhook, schedule and event triggers ignore the sample and always run the step for real.