#Extraction Fallback

extractPdf(...) is the high-level helper intended for OpenClaw-style model input.

import { extractPdf } from "clawpdf";

const result = await extractPdf("report.pdf", {
  mode: "auto",
  maxPages: 20,
  minTextChars: 200,
  image: {
    dpi: 96,
    maxPixels: 4_000_000,
    maxDimension: 10_000,
    forms: true,
  },
});

Flow for mode: "auto":

1. Extract text from selected pages.
2. Return text only when text length reaches minTextChars.
3. Otherwise render selected pages as compressed PNG images.
4. Stop rendering when image budget is exhausted.

#Modes

• auto: always extract text; render images only when text is short.
• text: extract text only.
• images: render images only.
• both: extract text and render images.

#Options

• pages: one-based pages to inspect.
• maxPages: finite positive maximum pages to inspect; the default 20 is ignored when pages is provided, but an explicit maxPages still caps that list.
• minTextChars: text threshold before image fallback, default 200.
• maxTextChars: text output cap, default 200_000.
• password: optional PDF user password.
• engine: optional PdfEngine for caller-owned reuse.
• image.dpi or image.scale: fallback render size, default dpi: 96.
• image.maxPixels: finite positive total rendered image pixel budget, default 4_000_000.
• image.maxDimension: finite positive maximum rendered PNG width or height, default 10_000.
• image.forms: render form widgets in fallback images, default true.

#Result

type ExtractResult = {
  text: string;
  images: Array<{
    page: number;
    width: number;
    height: number;
    bytes: Uint8Array;
    mimeType: "image/png";
  }>;
  pagesProcessed: number[];
  truncated: {
    text: boolean;
    images: boolean;
  };
};

Image bytes are raw PNG data. Use toMessageContent(result) or toDataUrls(result) when a transport needs base64.