Input nodes

Input nodes bring text into a flow. Every flow starts with one. You choose an input based on where your text is coming from.

Node	Use it when…
Text input	You want to paste or type text directly
JSON input	Your content comes as a structured JSON document
S3 input	You want to process a batch of files stored in Amazon S3 cloud storage

The text or document an input provides becomes the starting material that extractor nodes work on.

Choosing an input

Text Input is the quickest way to test a flow. The article is entered when you run it, so the same flow can be tried with different sample stories.

JSON Input is useful when you have structured articles and want to keep fields such as headline, URL, author, publication date, and images alongside the text. Paste one article object, or upload up to 20 .json files by drag-and-drop or file picker. One file creates a single-item run; two or more create one processed item per file, similar to a small S3 batch without cloud storage.

S3 Input is designed for repeatable batches. It scans a bucket and prefix for article JSON files, creates one processed item per accepted file, and keeps an ingestion ledger so unchanged object versions are not processed again by later runs. Use Process files again only when a deliberate rerun should ignore that protection.

JSON format

JSON and S3 inputs expect each article as a JSON object. The only required field is text; the other fields are optional metadata that make processed items easier to identify and review.

{
  "publication": "Example Daily",
  "headline": "City council approves bridge repairs",
  "url": "https://example.com/news/bridge-repairs",
  "author": "Riley Chen",
  "pub_date": "2024-05-14",
  "updated": "2024-05-14T18:15:18.425000-05:00",
  "text": "The city council approved emergency repairs to the Riverside Bridge on Tuesday.\n\nMayor Jane Doe said the work will begin next month and is expected to take six weeks.\n\nThe project will be managed by the Department of Transportation.",
  "images": [
    {
      "id": "image:bridge-repairs-001",
      "url": "https://example.com/images/bridge-repairs.jpg",
      "caption": "The Riverside Bridge will close for repairs next month."
    }
  ],
  "entry_id": "https://example.com/news/bridge-repairs"
}

For JSON input, paste one object or upload files in this shape. For S3 input, store each article as a .json file with this shape, then point the node at the bucket and prefix that contain those files.

Identity and updates

The optional entry_id gives an article a stable source identity. A durable CMS URL or source-system ID works well. Stable identity helps Backfield recognize that a later run refers to an article it has seen before, rather than creating an unrelated duplicate.

For S3, object version information also controls whether a file is considered unchanged. Moving or rewriting source files can therefore affect what a later scan sees.

When an article is saved, Backfield uses the optional publication field as its outlet in search and API responses. If publication is omitted, the URL hostname is used instead.

Images

Each image may include a source ID, URL, and caption. Image-aware nodes can describe or embed these images, and the processed item keeps them with the article. The image file still needs to be reachable by the services performing that work.

Before a run

The input panel shows the fields required by the selected input. S3 also needs configured AWS integration credentials. Input nodes bring content into Agate; collecting stories from a CMS, feed, or scraper happens outside the flow.