PDF block

Configure the PDF block to extract text and pages from a PDF document, with the operations it supports and the output each one returns.

The PDF block reads the text of a PDF, saves a copy cut down to selected pages, or reports how many pages it has. It reads the text layer; it does not recognise text in scanned images. In a workflow definition it is a step with type: "action" and action set to pdf.readText, pdf.extractPages or pdf.getInfo. For the plain-language guide, see Read PDF documents.

When to use it

Pull the text of an invoice or statement so an AI Agent step can extract fields from it.
Keep only the remittance page of a long payment pack before sending it on.
Check the page count first, then read only the last pages.

For a scanned document, pass the file to an AI Agent step as a document instead.

Operations

`action`	Editor label	What it does
`pdf.readText`	Read text	Returns the text of each selected page, and all of it joined.
`pdf.extractPages`	Extract pages	Saves a PDF with only the selected pages, or returns it inline.
`pdf.getInfo`	Get info	Returns page count, size and whether the PDF is encrypted, without reading text.

Configuration

Set one source. When more than one is set, fileId wins.

Field (`config` key)	Editor label	Type	Used by	Meaning
`fileId`	PDF File	string	All	Id of a PDF in Files, such as `{{steps.save_bill.file.id}}`.
`url`	Or a Link	string	All	An `https` URL to fetch the PDF from. Goes through Loopfour's outbound network policy, which refuses plain `http` and private addresses.
`content`	Or Inline Content	string	All	Base64 PDF bytes from an earlier step. Keep this for small files.
`contentEncoding`	Content Encoding	string	All	`base64` (default) or `text`, for `content`.
`fileName`	Not shown	string	All	Name to report for a `url` or `content` source. Default `document.pdf`.
`pages`	Pages	string	`readText`, `extractPages`	1-based pages and ranges, such as `1-5, 8, 11-13`. A range past the end is shortened; a single page past the end is dropped. Empty means the whole document.
`firstPages`	First Pages	number	`readText`, `extractPages`	Pages to take from the front. Combines with the other page options.
`lastPages`	Last Pages	number	`readText`, `extractPages`	Pages to take from the end. Combines with the other page options.
`mergePages`	Merge Pages	boolean	`readText`	Reported back as `merged`. Both `pages` and `text` are always returned.
`output`	Return	string	`extractPages`	`file` (default) saves the PDF in Files and returns its id. `content` returns base64 in the step output.
`baseName`	Saved File Name	string	`extractPages`	Name to save under. A re-run with the same name replaces the file. Defaults to the source name plus the pages kept.
`folderPath`	Folder Path	string	`extractPages`	Folder to save in, such as `Remittances/2026-11`. Defaults to the source file's folder.
`folderId`	Folder ID	string	`extractPages`	An existing folder. With `folderPath` too, the path is created inside it.
`maxSizeBytes`	Max Inline Bytes	number	`extractPages`	Cap for `output: "content"`. At most 14680064 (14 MiB).

Output

pdf.readText:

Field	Type	Meaning
`pages`	string array	Text of each selected page, in order.
`text`	string	All selected pages joined.
`pagesRead`	number array	1-based page numbers read.
`pageCount`	number	Pages in the document.
`merged`	boolean	The `mergePages` setting.
`fileName`	string	Name of the file read.
`sourceFileId`	string	The source file id, when the source was in Files.

pdf.extractPages:

Field	Type	Meaning
`fileId`, `filePath`, `fileName`	string	The saved PDF, for `output: "file"`. When no pages were dropped and no name or folder was set, `fileId` is the source file itself.
`content`, `contentEncoding`	string	Base64 bytes, for `output: "content"`.
`pagesKept`	number array	1-based pages kept. Empty when the result has every page.
`trimmed`	boolean	`true` when pages were dropped.
`pageCount`	number	Pages in the result.
`sourcePageCount`	number	Pages in the source.
`mimeType`, `sizeBytes`, `sourceFileId`		File facts.

pdf.getInfo: pageCount, fileName, sizeBytes, encrypted, sourceFileId.

Example

This workflow reads the first two pages of a bill saved in Files and asks an AI Agent step to extract the totals.

{
  "name": "Read bill totals",
  "trigger": { "type": "api" },
  "steps": [
    {
      "id": "read_bill",
      "type": "action",
      "name": "Read first pages",
      "action": "pdf.readText",
      "config": { "fileId": "{{input.fileId}}", "firstPages": 2 }
    },
    {
      "id": "extract_totals",
      "type": "agent",
      "name": "Extract totals",
      "config": {
        "prompt": "From this bill text, return the subtotal, tax and total.\n\n{{steps.read_bill.text}}",
        "outputSchema": {
          "type": "object",
          "properties": {
            "subtotal": { "type": "number" },
            "tax": { "type": "number" },
            "total": { "type": "number" }
          }
        }
      }
    }
  ]
}

Errors and limits

A source that is not a PDF fails the step. Sources over 96 MiB are refused.
readText reads at most 500 pages per run. A whole document over 500 pages fails; select pages instead.
Extracted text is capped at 10 MiB and the serialized result at 20 MiB.
Fetching a url and processing each have a 120-second limit.
extractPages with output: "content" fails when the result is larger than maxSizeBytes. Use output: "file" for large results.