dsh-deepread / site

# DeepRead 证据精读

DeepRead is an open-source evidence-first reading tool for AI agents.
DeepRead 将公众号文章、PDF、书籍和文档集整理成带原文证据、置信度、知识地图与复习问题的报告。

## Canonical identity

- Name: DeepRead 证据精读
- Package: dsh-deepread
- License: MIT
- Repository: https://github.com/xiehuan123/dsh-deepread
- npm: https://www.npmjs.com/package/dsh-deepread
- Website: https://xiehuan123.github.io/dsh-deepread/

## Install

- Agent Skill: npx skills@latest add xiehuan123/dsh-deepread
- DeepSeek Harness plugin: dsh plugin --profile web add dsh-deepread

## Main capabilities

- Read WeChat articles, text, Markdown, HTML and text-layer PDFs.
- Pair important claims with source evidence and locations.
- Separate author intent, source facts, reasonable inference and unverifiable content.
- Produce knowledge maps, data tables, recall questions and XMind-compatible outlines.
- Compare 2-10 documents for agreement, conflict and complementary evidence.
- Support quick, deep, map, feynman and book reading modes.

## Documentation and evidence

- Chinese README: https://github.com/xiehuan123/dsh-deepread/blob/main/README.zh.md
- English README: https://github.com/xiehuan123/dsh-deepread/blob/main/README.md
- Deep example: https://github.com/xiehuan123/dsh-deepread/blob/v1.0.0/examples/claude-code-token-optimization.md
- Knowledge map example: https://github.com/xiehuan123/dsh-deepread/blob/v1.0.0/examples/ad-fact-check-knowledge-map.md
- Architecture example: https://github.com/xiehuan123/dsh-deepread/blob/v1.0.0/examples/vivo-tauri-architecture.md

## Limits

- Scan-only PDFs should be OCR processed first.
- Anti-bot websites may require pasted source text.
- DeepRead exposes evidence and uncertainty but does not replace the reader's final judgment.