@dornick/parsers

File parser registry for Dornick — lazy VFS sub-mounts under /user/files/<id>/parsed/*.

Each parser materialises text or structured artifacts from an uploaded file. They register with a ParserRegistry; reads against /user/files/<id>/parsed/<sub> look up the matching parser and run it on demand, caching the result. WASM dependencies inside companion packages (@dornick/parsers-pdf, @dornick/parsers-office, @dornick/parsers-ocr, @dornick/parsers-ml, @dornick/parsers-audio) load only when first accessed.

Install

npm i @dornick/parsers

Capability-aware via a peer of @dornick/capabilities.

Built-in parsers

Three batteries-included parsers ship in this package:

| Parser | Matches | Produces | |---|---|---| | textParser | .txt, .md, .log, .yaml, anything text/* | text.txt | | jsonParser | .json, application/json | pretty.json, summary.txt | | csvParser | .csv, .tsv, text/csv | rows.jsonl, summary.txt |

The CSV parser handles RFC 4180-style quoting + escaped quotes + multi-line cells, without any runtime dependency.

Use

import { Dornick } from "@dornick/core";
import { ParserRegistry, BUILTIN_PARSERS } from "@dornick/parsers";

const parsers = new ParserRegistry(dornick.capabilities);
for (const p of BUILTIN_PARSERS) parsers.add(p);

// Pass via DornickOptions (planned wiring in @dornick/core)
new Dornick({ config, transport, parsers });

The model then reads /user/files/<id>/parsed/rows.jsonl (CSV), /parsed/text.txt (plain text), etc. Companion parser packages slot in identically:

import { pdfTextParser } from "@dornick/parsers-pdf";   // future package
import { mammothParser, xlsxParser } from "@dornick/parsers-office";
parsers.add(pdfTextParser);
parsers.add(mammothParser);
parsers.add(xlsxParser);

Define your own

import { defineSkill } from "@dornick/skills";   // unrelated — for context
import type { FileParser } from "@dornick/parsers";

export const exifParser: FileParser = {
  name: "exif",
  description: "Read EXIF metadata from JPEG/PNG uploads.",
  requires: ["wasm"],                          // capability gate
  matches: (file) => file.mimeType.startsWith("image/"),
  paths: () => ["exif.json"],
  parse: async (file, sub) => {
    if (sub !== "exif.json") throw new Error(`exif does not produce ${sub}`);
    const { extractExif } = await import("./_lazy/exif.js");
    return JSON.stringify(await extractExif(file.base64), null, 2);
  },
};

Two rules:

1. Be lazy — dynamic-import any heavy dependency inside parse(), not at module load.
2. Declare your gate — list every required BrowserCapabilities dot-path in requires. The registry filters you out on browsers that can't run you.

Capability gating

ParserRegistry.lookup and ParserRegistry.paths consult the registry's CapabilityRegistry. A parser declaring requires: ["webgpu", "crossOriginIsolated"] simply doesn't appear in paths() output or read() resolution on a browser without WebGPU. The model never sees an affordance that can't run.

Links

• Concept: Parsers
• Repo & root README

License

MIT — © Dipankar Sarkar