@openinspection/citera

Headless citation primitives for locating cited text and deriving highlight geometry.

import { citera, project } from "@openinspection/citera";

const cite = citera();

const result = await cite({
  text: "Revenue increased by 12% compared with the prior year.",
  pageIndex: 0,
  from: sourceTextModel,
});

const projected = project(result);

Citera is source-agnostic and viewer-agnostic. Consumers call a configured Citera engine with a citation request, then paint returned HighlightRegion geometry in their own viewer. A SourceTextModel works with core alone; raw PDF and image sources are handled by resolution packages.

Core normalizes supported from values before resolution stages run:

• SourceTextModel
• typed Blob
• URL
• explicit { kind, data } byte wrappers

Core loads and classifies sources, but parsing PDFs or recognizing images belongs to resolution packages.

Resolution stages

The core package includes text-model resolution. PDF parsing and OCR resolution live in separate packages and are exposed through the curated public seam.

import { citera, textResolution } from "@openinspection/citera";
import { pdfResolution } from "@openinspection/citera/resolutions/pdf";
import { ocrResolution } from "@openinspection/citera/resolutions/ocr";

const cite = citera({
  resolution: [
    pdfResolution(),
    textResolution(),
    ocrResolution({
      renderPage: async (page) => renderPageToImage(page),
    }),
  ],
});

const result = await cite({ text, pageIndex: 0, from: pdfFile });

The curated @openinspection/citera/resolutions/* subpaths keep heavier PDF and OCR dependencies out of the root entry point unless those resolution stages are imported.

Fixtures

fixtures/pdfs contains Huawei SUN2000-50KTL-M3 datasheets copied from the adjacent docs-solar workspace. fixtures/models contains a PDF.js-derived SourceTextModel for the English datasheet, and fixtures/citations contains stable citation inputs used by the test suite.