Skip to main content

​
Overview

Create workflows programmatically using the Kadoa SDK, CLI, MCP Server, or REST API:
  • Create workflows with natural language prompts
  • Use existing schemas or define custom ones
  • Configure monitoring and scheduling options

​
Prerequisites

Before you begin, you’ll need:
  • A Kadoa account
  • Your API key
  • For SDK: npm install @kadoa/node-sdk or yarn add @kadoa/node-sdk or uv add kadoa-sdk

​
Authentication

import { KadoaClient } from '@kadoa/node-sdk';

const client = new KadoaClient({
  apiKey: 'YOUR_API_KEY'
});

const status = await client.status();
console.log(status);
console.log(status.user);

​
Extraction Methods

Choose how you want to extract data from websites:

​
Auto-Detection

Auto-detect uses AI to detect and extract what’s on the page. If you’re using the REST API directly, auto-detection isn’t available and you need to pass a data schema.
// SDK: AI automatically detects and extracts data
const result = await client.extraction.run({
  urls: ["https://sandbox.kadoa.com/ecommerce"],
  name: "Auto Product Extraction",
  limit: 10,
});

console.log(result.data);

​
Custom Schema

Define exactly what fields you want to extract for precise control: 
const workflow = await client
  .extract({
    urls: ["https://sandbox.kadoa.com/ecommerce"],
    name: "Structured Product Extraction",
    extraction: (builder) =>
      builder
        .entity("Product")
        .field("title", "Product name", "STRING", {
          example: "iPhone 15 Pro",
        })
        .field("price", "Price in USD", "MONEY")
        .field("inStock", "Availability", "BOOLEAN")
        .field("rating", "Rating 1-5", "NUMBER")
        .field("releaseDate", "Launch date", "DATE"),
  })
  .create();

const result = await workflow.run({ limit: 10 });

// Use destructuring for cleaner access
const { data } = await result.fetchData({});
console.log(data);
Available Data Types: STRING, NUMBER, BOOLEAN, DATE, DATETIME, MONEY, IMAGE, LINK, OBJECT, ARRAY See all data types →

​
Classification

Automatically categorize content into predefined classes: 
const workflow = await client
  .extract({
    urls: ["https://sandbox.kadoa.com/news"],
    name: "Article Classifier",
    extraction: (builder) =>
      builder
        .entity("Article")
        .field("title", "Headline", "STRING", {
          example: "Tech Company Announces New Product",
        })
        .field("content", "Article text", "STRING", {
          example: "The article discusses the latest innovations...",
        })
        .classify("sentiment", "Content tone", [
          { title: "Positive", definition: "Optimistic tone" },
          { title: "Negative", definition: "Critical tone" },
          { title: "Neutral", definition: "Balanced tone" },
        ])
        .classify("category", "Article topic", [
          { title: "Technology", definition: "Tech news" },
          { title: "Business", definition: "Business news" },
          { title: "Politics", definition: "Political news" },
        ]),
  })
  .create();
//Note: 'limit' here is limiting number of extracted records not fetched
const result = await workflow.run({ limit: 10, variables: {} });
console.log(result.jobId);
const data = result.fetchData({ limit: 10 });
console.log(data);

​
Prompts

Every workflow is driven by a prompt: a plain-language description of what to extract and how to get there. The AI agent handles navigation, pagination, clicking, and forms automatically. For details, see Prompts.
Navigation is handled automatically when you provide a prompt.

​
Using a Prompt

For multi-step or complex extractions, write a prompt in plain language. The AI agent follows your prompt to navigate, interact, and extract: 
const workflow = await client
  .extract({
    urls: ["https://sandbox.kadoa.com/ecommerce"],
    name: "Product Catalog",
    userPrompt: "Extract all products. Click 'Next' to continue through pagination. For each product, open the detail page and extract the full description and specifications.",
    extraction: (builder) =>
      builder
        .entity("Product")
        .field("title", "Product name", "STRING", {
          example: "Wireless Headphones",
        })
        .field("price", "Product price", "MONEY")
        .field("description", "Full description", "STRING")
        .field("specifications", "Technical specs", "STRING"),
  })
  .create();

const result = await workflow.run({ limit: 50 });
Learn how to write effective prompts →
Complex multi-step extractions can take significantly longer to complete (usually around an hour). We recommend avoiding waiting for results synchronously.

​
Writing Prompts

Write clear, step-by-step prompts for complex extraction tasks. The AI agent follows your prompt autonomously, handling clicks, forms, pagination, and file downloads. Learn how to write effective prompts →

​
Create from a Template

If your team has templates set up, you can create a workflow and link it to a template. The template’s prompt, schema, and notification settings are applied to the workflow. 
> "Create a workflow from the 'Job Listing' template for https://example.com/careers"
Linking applies the template’s latest version. The template-managed parts become read-only on the workflow. See Templates for full details.

​
Next Steps