Workflow JSON

Building workflows with JSON

Every workflow in MultiverseAI is a JSON document. You can write one by hand, generate it from a script, keep it in version control, and paste it straight into the builder — or export what you built on the canvas and edit it as text. This page is the complete reference for that format.

Jump to quick start[ Format version 1.0 ]

Overview

Three ways to move JSON in and out of the builder — all of them use the format described on this page.

JSON editor

The { } button in the builder toolbar opens the current workflow as editable JSON. Edit it, validate live, then Apply to canvas.

Export

Downloads the workflow on the canvas as a .json file — including unsaved changes, and before the workflow has ever been saved.

Import

Reads a .json file back in. The document is validated before anything is written, so a malformed file is rejected with a specific reason.

A workflow is a directed graph: nodes are the steps, edges are the connections between them. Execution starts at a trigger node and follows the edges. That is the whole model — everything below is detail about how to spell it.

Quick start

The shortest workflow the builder will accept. Paste this into the JSON editor and press Apply to canvas.

{
  "nodes": [
    { "id": "start", "type": "trigger" },
    { "id": "reply", "type": "aiAgent" }
  ],
  "edges": [
    { "source": "start", "target": "reply" }
  ]
}

Two nodes and one connection. No positions, no data, no envelope — all of it is filled in for you.

Notice what is missing: there is no position, no data, no edge id, and no version wrapper. Every one of those is optional. The builder places the nodes down the canvas, fills each node's data with that type's defaults, and generates the edge id — telling you what it did in the warnings panel.

Write only the fields you actually want to control. Add detail as you need it.

Document shape

The editor and the importer accept three shapes. They are equivalent — use whichever is convenient.

1. The export envelope

This is what Export produces and what the editor shows you.

{
  "version": "1.0",
  "exportedAt": "2026-08-06T09:00:00.000Z",
  "workflow": {
    "name": "My Workflow",
    "description": "What this workflow does",
    "nodes": [ /* … */ ],
    "edges": [ /* … */ ]
  }
}

Field	Type	Required	Notes
`version`	string	No	Format version. A different value still imports, with a warning telling you to re-export.
`exportedAt`	string	No	ISO timestamp, written on export. Ignored on import.
`workflow.name`	string	No	Defaults to "Untitled Workflow". Applying JSON renames the open workflow.
`workflow.description`	string	No	Free text. Defaults to empty.
`workflow.nodes`	Node[]	Yes	The steps. May be empty, but the key must be present.
`workflow.edges`	Edge[]	No	The connections. Defaults to an empty list.

2. A bare workflow

Drop the envelope and put name, nodes and edges at the top level.

{
  "name": "My Workflow",
  "nodes": [ /* … */ ],
  "edges": [ /* … */ ]
}

3. A bare graph

Just nodes (and optionally edges). This is the form to generate from a script.

{
  "nodes": [ /* … */ ]
}

The node object

A node is one step. Its type decides what it does; its data configures it.

{
  "id": "summarise",
  "type": "aiAgent",
  "position": { "x": 250, "y": 350 },
  "data": {
    "label": "Summarise the posts",
    "model": "claude-3-5-haiku-20241022",
    "systemPrompt": "Summarise the input as five bullet points.",
    "temperature": 0.3,
    "maxTokens": 1024
  }
}

Field	Type	Required	Notes
`id`	string	Yes	Unique within the document. Edges reference it, so pick something readable — "summarise" beats "node_3".
`type`	string	Yes	One of the values in the catalog. An unrecognised type is an error, with a suggestion when it looks like a typo.
`position`	{ x: number, y: number }	No	Canvas coordinates in pixels; x grows right, y grows down. Omit it and nodes are stacked 150px apart at x = 250.
`data`	object	No	Type-specific configuration, merged over the type defaults. Anything you leave out keeps its default.
`data.label`	string	No	The name shown on the canvas. Defaults to the node type’s display name.
`style / hidden / parentId / extent`	various	No	Optional canvas presentation fields. Preserved on round-trip, safe to ignore.

Defaults are merged, not replaced

Supplying data does not wipe the rest of the configuration — your keys are layered on top of the type's defaults. Writing this:

{
  "nodes": [
    { "id": "ai", "type": "aiAgent" }
  ]
}

produces a fully configured node:

{
  "id": "ai",
  "type": "aiAgent",
  "position": { "x": 250, "y": 50 },
  "data": {
    "label": "AI Agent",
    "nodeLabel": "Process with AI",
    "model": "claude-3-5-haiku-20241022",
    "systemPrompt": "",
    "temperature": 0.7,
    "maxTokens": 2048
  }
}

So to change one setting, write one key. The catalog lists every default.

The edge object

An edge connects one node's output to another node's input. Output flows along it.

{
  "id": "e2",
  "source": "scrape",
  "target": "summarise",
  "sourceHandle": null,
  "targetHandle": null,
  "type": "default"
}

Field	Type	Required	Notes
`source`	string	Yes	The id of the node the data comes from. Must exist in this document.
`target`	string	Yes	The id of the node the data goes to. Must exist in this document.
`id`	string	No	Unique within the document. Generated for you if omitted.
`sourceHandle`	string \| null	No	Which output of the source node to leave from. Only meaningful for branching nodes.
`targetHandle`	string \| null	No	Which input of the target node to arrive at. Almost always omitted.
`type`	string	No	Edge rendering style. Defaults to "default".

An edge pointing at an id that does not exist is an error, not a silent drop — a typo in an id would otherwise quietly disconnect part of your workflow.

Branching

Most nodes have one output. Branching nodes have several, and sourceHandle picks which one an edge leaves from.

Node type	Valid sourceHandle	Notes
`condition`	"true", "false"	Two outputs — the comparison result.
`filter`	"pass", "fail"	Items that matched, and those that did not.
`switch`	"output-0", "output-1", …	One per output; `data.outputs` sets how many (default 3).
`parallel`	"branch-0", "branch-1", …	One per entry in `data.branches` (default 2).

{
  "edges": [
    { "id": "yes", "source": "check", "target": "approve", "sourceHandle": "true" },
    { "id": "no",  "source": "check", "target": "reject",  "sourceHandle": "false" }
  ]
}

Routing the two outputs of a condition node.

If you leave sourceHandle off an edge that comes from a branching node, the builder assigns handles in the order the edges appear — first edge "true", second "false", and so on — and warns you that it did. That is convenient while drafting and fragile afterwards: reordering the edges silently swaps the branches. Set the handle explicitly in anything you intend to keep.

Expressions

Any string in data can pull values from the running workflow using {{ … }}.

{
  "id": "email",
  "type": "sendEmail",
  "data": {
    "to": "{{$json.customer.email}}",
    "subject": "Your order {{$json.orderId}}",
    "body": "Processed at {{$now}} by workflow {{$workflow.name}}."
  }
}

Field	Type	Required	Notes
`{{$json.field}}`	any	No	A field from the current item coming into this node. The one you will use most.
`{{$items}} / {{$item(i)}}`	array / object	No	All incoming items, or the item at index i.
`{{$itemIndex}}`	number	No	Index of the item currently being processed.
`{{$node["id"].json.field}}`	any	No	Output of any earlier node by id — not just the one directly upstream.
`{{$workflow.id}} / {{$workflow.name}}`	string	No	The running workflow.
`{{$execution.id}} / {{$execution.mode}}`	string	No	This run. Mode is manual, trigger, webhook or integrated.
`{{$now}} / {{$today}}`	Date	No	Current timestamp, and midnight this morning.

Expressions are evaluated in a sandbox. Anything that reaches outside it — fetch, require, process, window, constructor and similar — is rejected, and expressions are capped at 2000 characters. For real logic, use a codeExecution node instead of a long expression.

Validation rules

The editor validates as you type. Errors block Apply; warnings tell you what was filled in on your behalf.

Errors — these must be fixed

The document is not valid JSON, or the top level is not an object.
nodes is missing, or nodes/edges is not an array.
A node has no id, or two nodes share one.
A node has no type, or a type the canvas cannot render.
A position is present but its x/y are not numbers.
data is present but is not an object.
An edge is missing source/target, points at an id that does not exist, or reuses an edge id.
The document has more than 500 nodes.

Structural rules

These match what the canvas refuses to let you draw by hand, so importing JSON cannot produce a graph you could not have built with the mouse.

No loops. The connections must flow one way. A cycle is reported with the path that closes it (a → b → c → a) — a cyclic workflow would never finish. To repeat work, use a loop node (loopOverItems, iterator) rather than an edge pointing backwards.
No self-connections. A node cannot feed its own input.
No trigger-to-trigger edges. Triggers start a workflow, so they cannot feed each other.
One connection per source/target pair. An exact duplicate is dropped with a warning. Two edges sharing a source and target but using different sourceHandle values are fine — that is how both branches of a condition converge on the same node.

Warnings — applied automatically

A missing position — the node is placed down the canvas.
A missing data — filled from the type defaults.
A missing edge id — generated from the endpoints.
A missing sourceHandle on a branching node — inferred by edge order.
A duplicate connection — the redundant edge is dropped.
A non-string data.label — the type default is used.
A version that is not 1.0.
No trigger node — the workflow imports, but nothing will start it.

Applying JSON replaces the entire canvas, and nothing is written to your account until you press Save. Export first if you want a copy of what is there now.

Node type catalog

Every value type accepts, with the data fields each one fills in by default. Omit a field and you get the default shown here. This list is generated from the running app, so it always matches what the builder will accept.

92 types

Actions

"action"Action

Perform various actions

data fields (default)

actionType = "message"messageTemplate = ""

AI & Logic

"aiAgent"AI Agent

AI-powered processing

data fields (default)

model = "google/gemini-3-flash-preview"systemPrompt = ""temperature = 0.7maxTokens = 2048

"condition"Condition

If/then logic

data fields (default)

variable = ""operator = "equals"value = ""

"filter"Filter

Filter data based on conditions

data fields (default)

filterType = "condition"condition = ""invertResult = false

"switch"Switch

Multi-branch routing logic

data fields (default)

mode = "rules"rules = []outputs = 3fallbackOutput = 0

AI/ML Enhancement

"aiEmbeddings"AI Embeddings

Convert text into high-dimensional vector embeddings

data fields (default)

model = "text-embedding-3-small"

"aiTransform"AI Transform

AI-powered data transformation and processing

data fields (default)

model = "google/gemini-3-flash-preview"instruction = ""

"contentSummarization"Content Summarization

Generate intelligent summaries from long-form content

data fields (default)

model = "google/gemini-3-flash-preview"maxLength = 200

"multiAIProvider"Multi-AI Provider

Orchestrate multiple AI providers for improved reliability

data fields (default)

providers = []strategy = "fallback"

"semanticSearch"Semantic Search

Intelligent semantic search with vector similarity

data fields (default)

query = ""topK = 5

"vectorDatabase"Vector Database

Store, search, and manage high-dimensional embeddings

data fields (default)

operation = "query"topK = 5

Canvas

"stickyNote"stickyNote

Added from the canvas, not the node palette.

data fields (default)

text = "Add a note..."color = "yellow"

Communication

"angelsend"AngelSend

Run any AngelSend email-outreach action: campaigns, steps, pause/resume/stop one recipient, merge vars, sender limits, opt-outs, webhooks

data fields (default)

action = "list-campaigns"paramsJson = "{}"

"chatInterface"Chat Interface

Send to Slack, Discord, Teams, Google Chat, Mattermost or Telegram

data fields (default)

platform = "slack"action = "send"channel = ""message = ""

"ftp"FTP/SFTP

Transfer files via FTP, SFTP, or FTPS

data fields (default)

host = ""username = ""password = ""operation = "list"

"graphql"GraphQL

Execute GraphQL queries, mutations, and subscriptions

data fields (default)

endpoint = ""query = ""variables = {}

"meetingNotetaker"Meeting Notetaker

Send an AI notetaker into a Google Meet, Zoom, Teams or Webex call, or read a meeting's summary, action items and transcript

data fields (default)

operation = "getMeeting"meetingId = "{{trigger.meeting_id}}"

"rssFeed"RSS Feed

Fetch and process RSS/Atom feeds

data fields (default)

feedUrl = ""maxItems = 20

"sendEmail"Send Email

Send emails via SMTP, Gmail, or other providers

data fields (default)

to = ""subject = ""body = ""

"slack"Slack

Post messages to Slack via your connected workspace or an incoming webhook

data fields (default)

message = ""channel = ""webhookUrl = ""

"ssh"SSH Command

Execute commands on remote servers via SSH

data fields (default)

host = ""username = ""command = ""

"webhookResponse"Webhook Response

Send custom HTTP responses back to webhook triggers

data fields (default)

statusCode = 200body = ""headers = {"Content-Type":"application/json"}

Data Processing

"codeExecution"Code Execution

Execute custom JavaScript code

data fields (default)

language = "javascript"code = "// return value becomes node output\nreturn in…

"compareDatasets"Compare Datasets

Compare two datasets and find differences

data fields (default)

compareKey = ""includeOnlyLeft = trueincludeOnlyRight = trueincludeBoth = true

"convertToFile"Convert to File

Convert data to various file formats

data fields (default)

targetType = "csv"filename = "output.csv"

"crypto"Crypto

Encryption, hashing, and cryptographic operations

data fields (default)

operation = "hash"algorithm = "sha256"

"databaseQuery"Database Query

Execute SQL queries with a visual editor and result preview

data fields (default)

connection = ""query = "SELECT 1;"

"dateTime"Date & Time

Date and time operations

data fields (default)

operation = "now"format = "YYYY-MM-DD HH:mm:ss"

"extractFromFile"Extract From File

Extract data from PDF, Excel, CSV files

data fields (default)

fileType = "pdf"source = ""

"jsonOperations"JSON Operations

Parse, transform, and manipulate JSON data

data fields (default)

operation = "parse"path = ""

"removeDuplicates"Remove Duplicates

Remove duplicate items from datasets

data fields (default)

keyField = ""scope = "all"

Data Sources

"gmail"Gmail

Read, search, send, and manage emails via Gmail API

data fields (default)

operation = "list"maxResults = 25

"googleCalendar"Google Calendar

Create, list and update events, add Meet links, check availability

data fields (default)

operation = "createEvent"

"googleChat"Google Chat

Post and manage messages in Google Chat spaces as you

data fields (default)

operation = "sendMessage"

"googleContacts"Google Contacts

Look up, add and update Google Contacts

data fields (default)

operation = "search"

"googleDocs"Google Docs

Create, read, and export Google Docs documents

data fields (default)

operation = "get"

"googleDrive"Google Drive

List, upload, download, and manage files in Google Drive

data fields (default)

operation = "list"

"googleForms"Google Forms

Build forms and read their responses

data fields (default)

operation = "listResponses"

"googleMeet"Google Meet

Create Google Meet links

data fields (default)

operation = "createSpace"

"googlePhotos"Google Photos

Browse albums and media items from Google Photos

data fields (default)

operation = "list"

"googleSearch"Google Search

Raw Google results (web, news or images) via Serper

data fields (default)

query = ""searchType = "general"numResults = 5

"googleSheets"Google Sheets

Read, write, append and look up rows in Google Sheets

data fields (default)

operation = "read"

"googleSlides"Google Slides

Create decks, add slides and fill {{placeholders}} in Google Slides

data fields (default)

operation = "create"

"googleTasks"Google Tasks

Create, list and complete Google Tasks

data fields (default)

operation = "createTask"

"promptInput"Prompt Input

Use saved prompt and response as input

data fields (default)

selectedPromptId = nullselectedPrompt = null

"tiktok"TikTok

Manage TikTok account, videos, and engagement

data fields (default)

operation = "getProfile"

"webscraper"Web Scraper

Extract content from websites using Firecrawl

data fields (default)

url = ""scrapeType = "markdown"onlyMainContent = truewaitFor = —

"websearch"Web Search

AI-powered web search with Perplexity

data fields (default)

query = ""model = "sonar"systemPrompt = ""searchRecencyFilter = ""domainFilter = ""maxTokens = —

"youtube"YouTube

Fetch video, channel, and search data from YouTube

data fields (default)

operation = "search"limit = 10

Flow Control

"aggregator"Aggregator

Collect and combine inputs

data fields (default)

aggregationType = "array"waitTimeout = 30

"delay"Delay

Wait for specified duration

data fields (default)

delayType = "fixed"delaySeconds = 1

"iterator"Iterator

Loop through arrays or repeat

data fields (default)

maxIterations = 10itemsSource = "input"

"loopOverItems"Loop Over Items

Process items in batches with loops

data fields (default)

batchSize = 10itemsSource = "input"

"parallel"Parallel

Execute branches in parallel

data fields (default)

branches = [{"id":"branch-0","label":"Branch 1","status":"…maxConcurrency = 2waitForAll = truefailFast = false

"subWorkflow"Sub-workflow

Execute another workflow

data fields (default)

workflowId = ""

"waitApproval"Wait for Approval

Pause workflow execution until human approval is given

data fields (default)

message = "Please review and approve."timeout = 3600

Integrations

"elevenlabs"ElevenLabs

Text to speech with your ElevenLabs voices, transcribe audio (Scribe), generate sound effects, list voices

data fields (default)

operation = "textToSpeech"text = ""modelId = "eleven_multilingual_v2"

"facebookAds"Facebook Ads

Run Meta Marketing API ops: campaigns, ad sets, ads, creatives, audiences, lead forms, and insights

data fields (default)

operation = "get-ad-accounts"limit = 25

"facebookFundraiser"Facebook Fundraiser

Manage fundraisers, pull donations, sync to CRM, and analyze fundraiser performance

data fields (default)

operation = "get-fundraisers"limit = 50

"facebookGroups"Facebook Groups

List groups, publish to groups, and read members & feeds via the Facebook Groups API

data fields (default)

operation = "facebookGroupGetGroups"limit = 25

"facebookPages"Facebook Pages

Publish posts, schedule content, reply to comments, and read Page insights on Facebook

data fields (default)

operation = "facebookPageGetPages"limit = 25

"heygen"HeyGen

AI avatar videos from a script or audio, and video translation with lip-sync

data fields (default)

operation = "heygenCreateVideo"aspectRatio = "16:9"resolution = "720p"engine = "avatar_iv"waitForCompletion = truemaxWaitSeconds = 600

"instagram"Instagram

Publish photos, reels & carousels, manage comments, send DMs, and read insights on Instagram

data fields (default)

operation = "instagramGetMedia"limit = 25

"integration"Integration

Connect to 300+ services and APIs with pre-built integrations

data fields (default)

provider = ""service = ""operation = ""authentication = {"type":"api-key","configured":false}executionStatus = "idle"

"jira"Jira

Create, update, search, and manage Jira issues

"linkedin"LinkedIn

Publish posts (text, images, video or a link), comment, react, delete posts, and read the profile of your connected LinkedIn member

data fields (default)

operation = "linkedinPublishPost"text = ""visibility = "PUBLIC"attachment = "NONE"

"messenger"Messenger

Send messages, list conversations, and read messages via the Facebook Messenger Platform

data fields (default)

operation = "messengerSendMessage"

"metaCatalog"Meta Catalog

Manage Meta commerce catalogs, products, and product sets across Facebook & Instagram

data fields (default)

operation = "metaGetCatalogs"limit = 25

"tavus"Tavus

Personalized replica videos and live AI video conversations

data fields (default)

operation = "tavusCreateVideo"waitForCompletion = truemaxWaitSeconds = 600maxMinutes = 10

"threads"Threads

Publish posts and carousels, reply, fetch threads/replies, and pull insights via the Threads API

data fields (default)

operation = "threadsGetPosts"limit = 25

"tiktokAds"TikTok Ads

Create and manage TikTok ad campaigns, audiences, and reporting

"tiktokAdTools"TikTok Ad Tools

Manage ad comments, blocked words, and automated optimization rules

"tiktokBrandSafety"TikTok Brand Safety

Manage video/profile exclusion lists and inventory filters

"tiktokCommerce"TikTok Commerce

Manage TikTok stores, catalogs, products, and store windows

"tiktokCreative"TikTok Creative

Manage TikTok creative assets, videos, images, templates, and discovery

"tiktokCreator"TikTok Creator

Manage TikTok creator partnerships, campaigns, and video linking

"tiktokDPACatalog"TikTok DPA Catalog

Manage TikTok product catalogs, feeds, product sets, and video packages

"tiktokFinance"TikTok Finance

Payment portfolios, credit lines, partner insights, and business plugin

"tiktokLeads"TikTok Leads

Manage TikTok lead generation forms, fields, and lead data

"tiktokMeasurement"TikTok Measurement

App/pixel event tracking, SDK config, deeplinks, and PPS surveys

"tiktokMentions"TikTok Mentions

Track brand mentions, hashtags, and comments on TikTok

"tiktokOfflineEvents"TikTok Offline Events

Track and manage offline conversion events for TikTok campaigns

"tiktokPixel"TikTok Pixel

Manage TikTok pixels, events, stats, and health reporting

"tiktokReporting"TikTok Reporting

Ad reporting, benchmarks, video performance, and Smart+ analytics

"tiktokRF"TikTok R&F

Manage TikTok Reach & Frequency campaigns, inventory, and contracts

"tiktokTCM"TikTok TCM

TikTok Creator Marketplace for campaigns, creator discovery, and reporting

"twitter"X (Twitter)

Post tweets, reply, search, manage lists, DMs, and more on X

data fields (default)

operation = "post"

"whatsappBusiness"WhatsApp Business

Send WhatsApp Business messages and approved templates via the Cloud API

data fields (default)

operation = "whatsappSendMessage"

Triggers

"errorTrigger"Error Trigger

Trigger on workflow errors

data fields (default)

catchAll = true

"manualTrigger"Manual Trigger

Manual trigger with forms

data fields (default)

formFields = []

"scheduleTrigger"Schedule Trigger

Trigger on schedule with cron

data fields (default)

cronExpression = "0 9 * * *"timezone = "UTC"

"trigger"Trigger

Start workflow with various triggers

data fields (default)

triggerType = "test"

Worked examples

Complete, valid documents you can paste in as-is.

A scheduled digest

Runs on a cron schedule, scrapes a page, summarises it with an AI agent, and emails the result.

{
  "version": "1.0",
  "workflow": {
    "name": "Daily Competitor Digest",
    "description": "Scrape a competitor's blog every morning, summarise it, and email the result.",
    "nodes": [
      {
        "id": "schedule",
        "type": "scheduleTrigger",
        "position": { "x": 250, "y": 50 },
        "data": {
          "label": "Every weekday at 09:00",
          "cronExpression": "0 9 * * 1-5",
          "timezone": "UTC"
        }
      },
      {
        "id": "scrape",
        "type": "webscraper",
        "position": { "x": 250, "y": 200 },
        "data": {
          "label": "Fetch the blog",
          "url": "https://example.com/blog",
          "scrapeType": "markdown",
          "onlyMainContent": true
        }
      },
      {
        "id": "summarise",
        "type": "aiAgent",
        "position": { "x": 250, "y": 350 },
        "data": {
          "label": "Summarise the posts",
          "model": "claude-3-5-haiku-20241022",
          "systemPrompt": "Summarise the following posts as five bullet points. Be specific about product changes.",
          "temperature": 0.3,
          "maxTokens": 1024
        }
      },
      {
        "id": "email",
        "type": "sendEmail",
        "position": { "x": 250, "y": 500 },
        "data": {
          "label": "Send the digest",
          "to": "team@example.com",
          "subject": "Competitor digest",
          "body": "{{$json.output}}"
        }
      }
    ],
    "edges": [
      { "id": "e1", "source": "schedule", "target": "scrape" },
      { "id": "e2", "source": "scrape", "target": "summarise" },
      { "id": "e3", "source": "summarise", "target": "email" }
    ]
  }
}

Branching on a value

A condition node routing urgent messages one way and everything else another, with explicit handles.

{
  "version": "1.0",
  "workflow": {
    "name": "Triage Incoming Message",
    "nodes": [
      {
        "id": "start",
        "type": "trigger",
        "position": { "x": 250, "y": 50 },
        "data": { "label": "When a message arrives", "triggerType": "message" }
      },
      {
        "id": "is_urgent",
        "type": "condition",
        "position": { "x": 250, "y": 200 },
        "data": {
          "label": "Is it urgent?",
          "variable": "{{$json.priority}}",
          "operator": "equals",
          "value": "urgent"
        }
      },
      {
        "id": "page_oncall",
        "type": "sendEmail",
        "position": { "x": 60, "y": 380 },
        "data": {
          "label": "Page on-call",
          "to": "oncall@example.com",
          "subject": "Urgent: {{$json.subject}}",
          "body": "{{$json.body}}"
        }
      },
      {
        "id": "queue_normal",
        "type": "action",
        "position": { "x": 440, "y": 380 },
        "data": {
          "label": "Add to the normal queue",
          "actionType": "queue"
        }
      }
    ],
    "edges": [
      { "id": "e1", "source": "start", "target": "is_urgent" },
      { "id": "e2", "source": "is_urgent", "target": "page_oncall", "sourceHandle": "true" },
      { "id": "e3", "source": "is_urgent", "target": "queue_normal", "sourceHandle": "false" }
    ]
  }
}

Batching over a list

Fan a list out in batches, call an API per batch, and collect the results back into one array. Written in the terse bare-graph form.

{
  "nodes": [
    { "id": "start",  "type": "trigger" },
    { "id": "fanout", "type": "loopOverItems", "data": { "batchSize": 25 } },
    { "id": "call",   "type": "graphql", "data": { "endpoint": "https://api.example.com/graphql" } },
    { "id": "collect","type": "aggregator", "data": { "aggregationType": "array" } }
  ],
  "edges": [
    { "source": "start",  "target": "fanout" },
    { "source": "fanout", "target": "call" },
    { "source": "call",   "target": "collect" }
  ]
}

Troubleshooting

The errors people hit most, and what they actually mean.

Field	Type	Required	Notes
`Unknown node type "…"`	error	No	The type is not in the catalog. Types are camelCase and case-sensitive: `aiAgent`, not `aiagent` or `AI Agent`.
`Edge references unknown target node`	error	No	The id in the edge does not match any node id. Often a rename that was applied to the node but not to the edges pointing at it.
`Duplicate node id`	error	No	Two nodes share an id — usually from copying a block and forgetting to rename it. Edges cannot tell them apart.
`Invalid JSON: …`	error	No	A syntax problem. Nearly always a trailing comma after the last item in an array or object; JSON does not allow them. The message carries the line and column.
`The connections form a loop: …`	error	No	An edge points back at a node earlier in the flow. The message names the exact path. Replace the backwards edge with a loop node (`loopOverItems`, `iterator`).
`"workflow" must be an object`	error	No	The `workflow` key is present but holds `null`, an array or a string. Give it a body, or drop it and use the bare shape.
`No trigger node`	warning	No	Nothing will start the workflow. Add a `trigger`, `scheduleTrigger`, `manualTrigger` or `errorTrigger`.
`Nodes overlap on the canvas`	—	No	Two nodes were given the same position. Either space them ~150px apart on y, or delete both positions and let them auto-place.

Still stuck?

Build a small version on the canvas, open the JSON editor, and read the JSON it produces. Comparing working output against what you wrote is the fastest way to find the difference.

Open the builder