dsh-docs: Fully Local Document Intelligence Plugin for DeepSeek Harness
sqhao-o/dsh-docs
Fully local document intelligence for DeepSeek Harness: parse PDF, Office, images, and scanned docs with offline OCR.
Enable DeepSeek Harness to parse PDFs, Office files, images, and scanned documents offline with OCR, no Docker or API keys required.
dsh plugin --profile web add github:sqhao-o/dsh-docs- Category
- Data & Knowledge
- Platform
- DSH-Plugin
- Author
- sqhao-o
- Distribution
- Plugin
dsh-docs Key Features
dsh-docs Repository
dsh-docs Repo Summary
dsh-docs is a document intelligence plugin (DSH plugin) for DeepSeek Harness, focused on fully local document parsing and OCR. It solves the problem of processing PDFs, Office files, images, and scanned documents in offline environments, without Docker, HTTP services, or API keys, ensuring no document ever leaves your disk. Core capabilities include converting PDF, Word, Excel, PowerPoint, Markdown, HTML, CSV, and other formats into clean Markdown, plain text, or structured JSON, along with an offline OCR pipeline (supporting English and Simplified Chinese) for scanned pages and images.
What document formats and OCR capabilities does the dsh-docs DSH plugin support?
dsh-docs supports PDF, DOCX, XLSX, PPTX, Markdown, HTML, CSV, and plain text, as well as local OCR for PNG, JPEG, TIFF, WebP, and scanned PDFs. It ships a pinned Python + Xberg runtime with offline Tesseract language data (English and Simplified Chinese), delivering complete PDF/Office/OCR coverage on Windows x64 out of the box. For non-OCR scenarios, the native Xberg Node binding serves as a lightweight fallback on any platform. All file reads are confined to folders you explicitly authorize, ensuring security.
How do I install and configure dsh-docs?
Installation requires no local checkout or build toolchain. Simply paste a prompt into a DSH session (e.g., dsh web), and the Harness agent will install the npm package, download the offline OCR runtime, and configure the plugin automatically. The only prerequisite is a working dsh CLI on Node ^22.19 or >= 24. Alternatively, you can run the installation command manually:
dsh plugin --profile web add github:sqhao-o/dsh-docs
On Windows x64, you also need to download the prebuilt offline OCR runtime and edit cordis.patch.yml to add the plugin configuration (e.g., engine: python, runtimeDir, defaultOcr: true). On other platforms, use engine: node and disable OCR. After configuration, restart dsh web, then check the engine with dshdoc_health and parse files with dshdoc_extract.
Which DeepSeek Harness versions and platforms are compatible with dsh-docs?
The plugin is maintained by Sqhao-O, licensed under MIT, and last updated in 2026-08. It is built with TypeScript and depends on official packages like @deepseek-ai/dsh-tools and @deepseek-ai/dsh-llm, ensuring seamless integration with the DeepSeek Harness framework. Platform-wise, Windows x64 supports full functionality (including OCR), while other platforms can use the Node engine for non-OCR parsing.
What are the test coverage and considerations for dsh-docs?
The integration test suite generates binary documents outside the repository and verifies parsing for PDF, DOCX, XLSX, PPTX, PNG OCR, and scanned-PDF OCR. For other Xberg-supported input formats, it is recommended to test against your own corpus before enabling in production. Additionally, note that the published package and plugin ID use the dsh-doc spelling, and tools use the dshdoc_* prefix, differing from the initial dsh-docling / docling_* release.