dsh-docs DSH plugin

dsh-docs is a DeepSeek Harness plugin that provides fully local document parsing and OCR capabilities. It supports PDF, Office documents, Markdown, HTML, CSV, images, and scanned PDFs, converting them to Markdown, plain text, or JSON. The plugin ships a self-contained Python runtime with offline Tesseract OCR for Windows x64, and a Node.js fallback for other platforms, with no HTTP service, Docker, or API keys required.

developer-tools skills-workflows

How to install the dsh-docs DSH plugin

dsh plugin --profile web add dsh-doc

Copying does not run this command. Review the repository and version before installing the dsh-docs DSH plugin.

dsh-docs DSH plugin data source

dsh-docs DSH plugin snapshot date: Aug 16, 2026

discovered

What the dsh-docs DSH plugin can do

Parse PDF, DOCX, XLSX, PPTX, Markdown, HTML, CSV, and plain text files.
Offline OCR for PNG, JPEG, TIFF, WebP, and scanned PDFs using a bundled Tesseract runtime (English and Simplified Chinese).
Output as Markdown, plain text, or structured JSON, configurable per request.
Security model: path realpath checking, byte snapshot before parsing, no remote URL fetching, and results bounded by maxOutputChars.
Self-contained embedded Python runtime (CPython 3.11.9, Xberg, Tesseract) for Windows x64; Node.js fallback for other platforms.
Configured through cordis.patch.yml with engine, runtimeDir, OCR languages, allowedLocalRoots, and more.

Where the dsh-docs DSH plugin fits

Extract tables from a local Excel file and ask the model to analyze them.
Read a scanned PDF invoice via offline OCR and get the text in plain text format.
Convert a batch of Word documents to Markdown for further processing in a DSH workflow.
Parse a PDF report and retrieve specific sections (e.g., risks, financials) via the DSH agent.
Enable local document understanding without sending any file to an external service.

Who the dsh-docs DSH plugin is for

DSH users who need to process local documents privately and securely.
Windows x64 users who want a complete offline OCR pipeline without external dependencies.
Developers building DSH-based workflows that require structured document extraction.

dsh-docs DSH plugin limitations

Full offline OCR (with bundled Tesseract) is only supported on Windows x64; other platforms must use the Node.js engine with OCR disabled or manually provide tessdataPath.
The plugin does not support fetching documents from URLs; remote files must be downloaded to an allowed local root before parsing.
Maximum output length is capped at 32000 characters by default (configurable via maxOutputChars).
The plugin is at an early stage; users should test unsupported Xberg input formats against their own corpus before production use.

dsh-docs DSH plugin: from the repository README

Quoted from the Sqhao-O/dsh-docs README, the upstream source of the dsh-docs DSH plugin. Copyright remains with the original authors.

**dsh-docs** gives your DeepSeek Harness agent real document intelligence — entirely on your own machine. Hand it a PDF, Word, Excel, or PowerPoint file and get back clean Markdown, plain text, or structured JSON; hand it a scanned page or image and a fully offline OCR pipeline reads it for you. No Docker, no HTTP service, no API keys, and no document ever leaves your disk. It ships a pinned, self-contained Python + [Xberg](https://github.com/xberg-io/xberg) runtime with offline Tesseract language data (English and Simplified Chinese), delivering complete PDF/Office/OCR coverage on Windows x64 out of the box. The native Xberg Node binding serves as a lightweight non-OCR fallback on any platform, and every file read stays confined to folders you explicitly authorize. The published package and plugin id use the `dsh-doc` spelling and the tools use `dshdoc_*`; they were renamed from the initial `dsh-docling` / `docling_*` release. ## One-prompt install No local checkout or build toolchain is needed. Paste the following prompt into a running DSH session (for example `dsh web`) in your own project folder. The Harness agent installs the published npm package, downloads the pinned off

Read the full READMERepository license: MIT

dsh-docs DSH plugin questions

How do I install the dsh-docs plugin?

Run `dsh plugin --profile web add dsh-doc` in a DSH session. The Harness agent will install the published npm package and configure the plugin. After installation, you need to restart `dsh web` before using the tools like `dshdoc_health` and `dshdoc_extract`.

Does the plugin support OCR on Linux or macOS?

On platforms other than Windows x64, the plugin uses the Node.js engine which does not include a bundled Tesseract runtime. You can enable OCR by setting `tessdataPath` to a local directory containing the required `.traineddata` files, but no automatic download is provided. For full offline OCR, Windows x64 is recommended.

Can I parse a file from a URL?

No. The plugin's `dshdoc_convert_url` tool is a stub that returns `UNSUPPORTED_URL`. You must download the file to an allowed local root (e.g., your session workspace or a directory listed in `allowedLocalRoots`) and then parse it with `dshdoc_extract` or `dshdoc_convert_file`.

How do I configure the allowed file size for parsing?

Set `maxFileBytes` in the plugin's config inside `cordis.patch.yml`. The default is 52428800 (50 MB). If a file exceeds this limit, the plugin will refuse to parse it.

What types of files can the plugin parse?

It supports PDF, DOCX, XLSX, PPTX, Markdown, HTML, CSV, plain text, PNG, JPEG, TIFF, WebP, and scanned PDFs. For scanned PDFs, OCR is applied if the `defaultOcr` option is enabled and the runtime provides Tesseract data.