Action Nodes

Action and flow nodes do the work between trigger and result: fetch data, run code, wait, branch, loop and call other workflows. Each node has inputs, which you type in or fill with variables, and named outputs that later nodes can use.

Settings Every Node Shares

Action, integration and agent nodes have two settings at the bottom of their panel:

• Error handling: Stop workflow (default) ends the run when the node fails. Use error output activates the node's error output instead, so you can connect a fallback path.
• Approval required: the run pauses before this node until someone approves it. See Approvals.

Actions

HTTP Request

Sends an HTTP request and returns the response.

• With Auto (text/JSON), a binary response (a PDF, an image, a ZIP) fails the node with a message to switch the response type to File (binary). A file is stored up to 100 MB.
• Sending the token of a Get Auth Credentials node: see Get Auth Credentials.

Web Search

Searches the web.

Delay

Pauses the run for a duration or until a point in time, then continues. Nothing keeps running while it waits.

If Resume At is already past, the run continues at once. A single Delay can wait at most 90 days. A longer wait fails the node.

RSS Feed

Fetches and parses an RSS or Atom feed.

Code

Runs JavaScript. Use it for filtering, sorting, string handling, calculations and other logic that needs no AI.

• Your code must return a value. That value becomes result.

• Outputs of earlier nodes are available as global variables under their variable name, for example formTrigger.outputs.formData. Do not use the {{…}} syntax here.

• async and await work, and fetch is available for HTTP calls.

• fetchAs('jira', '/rest/api/3/myself') calls a connected service with its credentials, exactly like the Authenticated Request node, and returns a response with status, ok, headers, text(), json() and bytes(). It uses a connection that another node of the same workflow already uses. If the workflow uses several connections of one service, name one: fetchAs('jira@<site or account of the connection>', …). The credentials never reach your code. A redirect to another host, as a download link answers with, is followed without the credentials; await res.bytes() then holds the file. Responses are limited to 10 MB.

• dg.log(), dg.debug(), dg.warn() and dg.error() write to the node's log in the execution history. console.log() works too.

• Besides standard JavaScript, these helpers are built in:

  • atob() and btoa() for Base64 (Latin-1 text only, as in the browser), and Buffer for bytes and other encodings, for example Buffer.from(text, 'utf8').toString('base64') or Buffer.from(b64, 'base64').toString('utf8').
  • TextEncoder and TextDecoder, including the options fatal and ignoreBOM.
  • JSZip to read and write ZIP files. Word, Excel and PowerPoint files are ZIP files too, so you can read the text of a downloaded document: const zip = await JSZip.loadAsync(await response.bytes()); const xml = await zip.file('word/document.xml').async('string');. Password-protected ZIP files cannot be opened.
  • setTimeout() and clearTimeout(), for example to wait between two requests. A timer has to fire within 60 seconds of the start, and at most 1,000 timers can be pending at once.
  • require('buffer') and require('jszip') return the same libraries.

• The code runs in an isolated sandbox. There is no file system, no npm packages and no import; require() knows only the two names above. FormData and Blob are not available.

• Limits per run:

  • 60 seconds of computing time, and 5 minutes in total including the time spent waiting for requests and timers.
  • 512 MB of memory.
  • A result of at most 10 MB.
  • The first 1,000 log lines or 1 MB of log text are kept.

  Over the time, memory or result limit, the run stops with an error that names the limit. Log lines beyond the limit are dropped with a notice.

const items = formTrigger.outputs.formData.items;
return items.filter((i) => i.status === 'open').map((i) => i.title);

Image Generation

Generates an image from a text prompt.

Data Tables

The sub-view Integrations → Data Tables in the node picker holds four nodes: Data Table: Get Rows, Insert Row, Update Rows and Delete Rows. Their inputs and outputs, and the three further nodes of the Data Table integration, are documented on the Data Table integration page.

Secret Stores

The sub-view Integrations → Secret Stores reads values from your Secret Stores:

• Get All Secrets: input Secret Store, output variables, an object of all keys and values
• Get Single Secret: inputs Secret Store and Variable Key, output value

Later nodes reference these outputs like any other variable, for example {{getAllSecrets.outputs.variables.API_KEY}}. Values marked as secret are passed on as references and only resolved when the next node runs, so the node outputs and the execution history show a reference instead of the value.

Flow

Condition

Routes the run into one of several branches. Each branch has a label and a condition. The first branch whose condition is true runs.

A condition is a JavaScript-like expression with variables:

{{webSearch.outputs.results.length}} > 0
{{httpRequest.outputs.statusCode}} === 200
{{aiAgent.outputs.result.sentiment}} !== "negative"
{{code.outputs.result.flag}} === true

Write the variable without quotes. Text values are compared as text, with the comparison value in quotes ("negative"); numbers and booleans without quotes (200, true). Text that is exactly a number, true, false or null is compared as that value, so "200" equals 200. A variable inside a regular expression (/…/) is not supported; to search text, use .includes("…").

Instead of an expression you can write the condition in plain language with the AI Prompt mode; an AI model then decides whether it holds. That costs tokens.

There is no automatic default branch. To catch everything else, add a last branch with the expression true.

Loop

Adding a loop places a Loop Start and a Loop End node. The nodes between them run once per item.

Loop Start takes an Array and provides per iteration:

Loop End takes an Iteration Result, the value to collect from each pass, and returns:

Subworkflow

Calls another workflow that starts with a Subworkflow Trigger and waits for it to finish. Choose the workflow and fill in the parameters its trigger defines.

The called workflow must be deployed, active and in the same workspace. Calls can be nested up to 10 levels deep. If the subworkflow pauses for an approval, the calling run pauses too.

The node's outputs contain the outputs of the subworkflow's nodes, grouped by node, for example {{subworkflow.outputs.<node>.<field>}}. The Variables panel shows the exact paths. Costs inside the subworkflow count toward the calling run.