Node Reference

Complete configuration reference for every node type. Each section includes the config schema, behavior description, input/output shapes, and an example.

Input Nodes

Input nodes are triggers that start process executions.

Gmail Input

Polls a Gmail inbox for new emails matching filters.

Type: inputNode with input type gmail

Config

Output Data

{
  "id": "msg_abc123",
  "thread_id": "thread_xyz789",
  "sender": "buyer@example.com",
  "recipients": ["orders@yourcompany.com"],
  "cc": [],
  "subject": "Steel Order Request",
  "body": "Plain text email body...",
  "body_html": "<p>HTML email body...</p>",
  "date": 1712937600,
  "receiving_account": "orders@yourcompany.com"
}

If includeThread: true, adds:

{
  "thread": [
    {
      "subject": "Re: Steel Order Request",
      "sender": "orders@yourcompany.com",
      "recipients": ["buyer@example.com"],
      "body": "Previous reply...",
      "date": 1712851200
    }
  ]
}

Output Files

Attachments are uploaded and available as files:

[
  {
    "name": "spec-sheet.pdf",
    "url": "https://storage.example.com/attachments/msg_abc123-0-spec-sheet.pdf",
    "size": 245760
  }
]

Polling Behavior

• Polls every pollingInterval seconds
• Uses exponential backoff when no emails found (up to 5 minutes)
• Resets to pollingInterval when emails arrive
• Query filter: in:inbox after:{last_poll_timestamp}

Outlook Input

Polls an Outlook inbox for new emails. Configuration mirrors Gmail Input with Outlook-specific auth.

Type: inputNode with input type outlook

Config and output are structurally identical to Gmail Input, using Outlook OAuth credentials instead.

API Input (Webhook)

Exposes an HTTP endpoint that triggers the process when called.

Type: inputNode with input type api

Config

Webhook URL

When the process is started, the webhook becomes available at:

POST /api/v1/processes/{processId}/webhook

Authentication: API key in the X-API-Key header or Authorization: Bearer {key}.

Input Data

The POST body becomes the node's output data. For JSON:

{
  "customer": "John Smith",
  "order_details": { ... }
}

For multipart form data, files are uploaded and available as file references.

Synchronous Response (Return to Client)

If the process has an output node of type Return to Client, the webhook blocks until execution completes and returns the output node's data as the HTTP response. This enables request-response workflows.

Scheduled Input

Triggers on a cron schedule.

Type: inputNode with input type scheduled

Config

Cron Expression Format

┌───────── minute (0-59)
│ ┌───────── hour (0-23)
│ │ ┌───────── day of month (1-31)
│ │ │ ┌───────── month (1-12)
│ │ │ │ ┌───────── day of week (0-6, 0=Sunday)
│ │ │ │ │
* * * * *

Examples:

• 0 9 * * 1-5 -- 9 AM every weekday
• */15 * * * * -- every 15 minutes
• 0 0 1 * * -- midnight on the first of every month

The UI includes an AI-assisted generator: describe your schedule in plain English and it generates the cron expression.

Output Data

{
  "triggered_at": "2026-04-12T09:00:00Z",
  "cron_expression": "0 9 * * 1-5"
}

Plus whatever is in the body config field.

Output Nodes

Output nodes deliver the final results of a process execution.

Gmail Output

Sends or replies to a Gmail message.

Type: outputNode with output type gmail

Config

Reply Mode

To reply to the original email thread (common pattern):

Thread ID:  {{input["Email Input"]["thread_id"]}}
Reply Mode: reply
Subject:    Re: {{input["Email Input"]["subject"]}}

The node fetches the original email, extracts the sender, and sends the reply to that sender within the same thread.

File Attachments

Reference files from upstream nodes:

input["files"]["Email Input"]        -- all files from Email Input
input["files"]["Email Input"][0]     -- first file only
input["files"]["OCR Node"][0]        -- first file from OCR Node

Output Data

{
  "message_id": "msg_def456",
  "thread_id": "thread_xyz789",
  "provider": "gmail",
  "success": true,
  "recipients": ["buyer@example.com"],
  "subject": "Re: Steel Order Request"
}

Outlook Output

Sends or replies to an Outlook message. Configuration mirrors Gmail Output with Outlook-specific auth.

Return to Client

Returns data synchronously to the webhook caller.

Type: outputNode with output type return_to_client

When a process with this output type is triggered via webhook, the HTTP response blocks until execution completes, then returns this node's input data as the response body.

Config

All upstream data available to this node is returned as the HTTP response.

Console Output

Logs output data for debugging. No external delivery.

Type: outputNode with output type console

Useful during development. The output data appears in the execution history.

Processing Nodes

Schematization

Extracts structured data from unstructured input using an LLM and a schema definition.

Type: schematizationNode

Config

Targeted Extraction

For fields that are difficult to extract reliably, configure targeted extraction:

{
  "targetedFields": [
    {
      "dataBlockId": "uuid",
      "fieldName": "total_amount",
      "description": "Look for the grand total at the bottom of the invoice",
      "maxAttempts": 3,
      "judgeModel": "gpt-4o",
      "hintImages": ["file-uuid"]
    }
  ]
}

The engine runs multiple extraction attempts at increasing temperatures and uses a judge model to select the best candidate.

Output Data

One key per selected data block:

{
  "Order Request": {
    "customer_name": "John Smith",
    "weight_lbs": 5000,
    "steel_type": "carbon"
  },
  "Quote": {
    "estimated_delivery": "10-14 business days"
  },
  "_metadata": {
    "llm": {
      "model": "gpt-4o",
      "prompt_tokens": 1234,
      "completion_tokens": 567
    }
  }
}

How It Works

1. Extracts text from input (plain text or OCR on files)
2. Optionally runs preprocessing function
3. Separates data blocks into solo (no edges) and coupled (with schema edges) groups
4. Solo blocks: extracted independently in parallel via LLM structured completion
5. Coupled blocks: root selected, then coupling functions derive related blocks
6. Field names remapped to match schema definitions
7. Targeted extraction runs for configured difficult fields

ML Model

Sends a prompt to an LLM and returns the response.

Type: mlModelNode

Config

How It Works

1. Input data is converted to text via ExtractTextFromInputs()
2. System prompt is sent as the system message
3. Extracted input text is sent as the user message
4. File attachments are included with the user message
5. LLM response is parsed -- if valid JSON, fields are accessible individually

Output Data

If the LLM returns valid JSON:

{
  "summary": "The order is for 5000 lbs of carbon steel",
  "sentiment": "neutral",
  "_metadata": { "llm": { "model": "gpt-4o", "duration_ms": 2300 } }
}

If the LLM returns plain text:

{
  "output": "The order is for 5000 lbs of carbon steel...",
  "_metadata": { ... }
}

Script

Executes Python code in a sandboxed environment.

Type: scriptNode

Config

Sandbox

The Python environment is sandboxed with restricted access to system resources. Network and filesystem access are disabled.

• Script timeout: 5 minutes

Accessing Input Data

All upstream data is available as the global input_data dict:

# Access a field from an upstream node
weight = input_data["Extract Order"]["Order Request"]["weight_lbs"]

# Access all input data
all_data = input_data

# Access text data
text = input_data["text"]

Returning Data

Single output:

return {"price": 2250.00, "delivery": "10 days"}

Multiple output ports:

return {
    "approved": {"status": "approved", "amount": 2250},
    "rejected": None  # this port gets no data
}

Output Data

{
  "output": {
    "price": 2250.00,
    "delivery": "10 days"
  }
}

Or for schematized output, the data is keyed by the data block name.

Conditional

Branches execution based on conditions or switch-case matching.

Type: conditionalNode

See Flow Control for detailed documentation.

Loop

Iterates a subgraph over an array of items or a set of files.

Type: loopNode

See Flow Control for detailed documentation.

Wait

Pauses execution for a configurable duration.

Type: waitNode

Config

The execution timeout is paused during wait nodes, so waits do not count against the overall timeout.

Output Data

Passes through all input data unchanged.

Database

Queries a connected SQL database.

Type: databaseNode

Config

Output Data

{
  "rows": [
    { "id": 1, "name": "Carbon Steel", "price_per_lb": 0.45 },
    { "id": 2, "name": "Stainless Steel", "price_per_lb": 1.20 }
  ],
  "row_count": 2
}

SQL

Executes raw SQL queries. Similar to Database node but with more direct control.

Type: sqlNode

OCR

Extracts text from images and PDF files.

Type: ocrNode

Config

Output Data

{
  "text": "Extracted text from the document...",
  "pages": [
    { "page": 1, "text": "Page 1 content..." },
    { "page": 2, "text": "Page 2 content..." }
  ],
  "_metadata": {
    "ocr": {
      "method": "ocr",
      "duration_ms": 3400,
      "pages_processed": 2
    }
  }
}

Web Automation

Controls a headless browser to interact with web pages.

Type: webAutomationNode

Config

Actions include: navigate to URL, click element, type text, extract text, screenshot, wait for element, etc.

Integration Nodes

Type: integrationsNode

Integration nodes connect to external services. The integration type is specified in the config.

API Fetch

Makes HTTP requests to any REST API.

Config

Example: POST with Templates

{
  "url": "https://api.crm.com/contacts",
  "method": "POST",
  "contentType": "application/json",
  "authType": "bearer",
  "dataTemplate": "{\"name\": \"{{input[\"Extract Order\"][\"Order Request\"][\"customer_name\"]}}\" , \"email\": \"{{input[\"Extract Order\"][\"Order Request\"][\"customer_email\"]}}\"}"
}

Output Data

{
  "status_code": 200,
  "body": { ... },
  "headers": { ... }
}

SendGrid

Sends transactional email via SendGrid API.

Config

SFTP

Uploads or downloads files via SFTP.

Config