Tools & Capabilities

dsh-document-evidence

Anduin9527/dsh-document-evidence · ★ 0

Anduin9527

Native PDF library for DeepSeek Harness Web: bounded evidence retrieval, page images and source citation highlighting.

License
AGPL-3.0
Added
2026-10-30
Maintenance
active
Install kind
github
npm
dsh-document-evidence
Manifest details
Manifest status
Declared
Manifest path
package.json#dsh
Manifest checked
2026-10-30T09:38:35.944Z
Lifecycle scripts
None found

Static manifest checks report declarations only; they are not compatibility or security review.

安装

dsh plugin --profile web add 'github:Anduin9527/dsh-document-evidence#d7a4dd35f9df9f83b79894c5df159307c372df81'

GitHub 源码安装:首次需按提示配置 allowBuilds 构建授权后重试

For install and environment setup, plus plugin development tutorials, see DSH community docs ↗

README

Source README · body: en · Read Chinese README

Table of contents
  1. Features
  2. Install
  3. Use
  4. Data and models
  5. More
  6. License

Ask questions about your PDFs and follow each citation back to the source page, right inside dsh Web.

中文 · Configuration · Changelog

Citation opening the original PDF page beside the conversation

Actual UI capture using a sample manual and a local protocol test model.

Features

  • A project PDF library shared across sessions, with a separate document selection for each conversation.
  • Background indexing with progress, page coverage, stop and resume controls.
  • Source retrieval, neighboring-page reading and clickable page citations.
  • Page images for charts and scanned content through your existing dsh vision model.
  • Native tool cards showing candidates, excerpts, timing and status.

Install

Requires dsh Web, Node.js 22.19+ on the 22.x line or 24+, and Python 3.10+ with venv. Verified with @deepseek-ai/dsh@0.1.3-alpha.2 on macOS arm64.

dsh plugin --profile web add dsh-document-evidence@beta
dsh plugin --profile web exec dsh-document-evidence-setup
dsh web

Restart dsh if it is already running. Setup creates a dedicated Python environment and installs PyMuPDF. The plugin uses your existing dsh model configuration; no additional API key is needed. Image reading requires a model with image input support.

Use

  1. Upload a PDF in a project conversation and open 资料库 (Library).
  2. Confirm that it should be added and indexed, or choose a project PDF.
  3. Select the documents to search and ask your question.
  4. Click a citation to open the original page and highlight its text.

The plugin instructs the agent to include clickable sources in PDF-based answers automatically; you do not need to ask for citations.

Try: “Compare revenue for the same year across these two reports. Check the units.”

The library supports filename search, document selection and index progress. Page citations support navigation and zoom; scanned pages use page-level navigation when no text coordinates are available.

Data and models

PDF snapshots and indexes stay local. Summaries, visual page cards and answers send the required excerpts or page images to your configured dsh model service and consume its usage. The default data directory is ~/.dsh/data/document-evidence, or under DSH_HOME when set.

More

Configuration and development · Performance measurements · Issues · Contributing

License

AGPL-3.0-only. See third-party notices.