{
  "id": 1306413,
  "title": "I Built DeepRead: An Evidence-First Reading Skill for Articles, Books, and PDFs",
  "url": "https://urgent.news/2026/08/16/i-built-deepread-an-evidence-first-reading-skill-for-articles-books",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-16T16:31:18.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/_312a710fe7837e8fcea442/i-built-deepread-an-evidence-first-reading-skill-for-articles-books-and-pdfs-45e8"
  },
  "original_language": "en",
  "account": "I recently developed an open-source project named DeepRead to address a recurring issue with AI reading tools. While these tools can summarize documents swiftly, they frequently blur the distinction between the original author's assertions, the evidence from the source, and the AI's own deductions. Moreover, when dealing with lengthy documents, crucial details such as source locations and context surrounding numerical claims may be lost. DeepRead transforms articles, books, PDFs, and document collections into structured reports centered around claims, evidence, data, and connections. Notably, the report links significant claims to their supporting evidence and source locations. Should the source lack evidence, the report explicitly states this instead of filling in the gap. DeepRead offers five reading modes: quick, deep, map, feynman, and book. Quick mode delivers the main idea and key takeaways; deep mode examines the core claim and argument structure; map mode generates a knowledge map with confidence levels, evidence pairs, data tables, relationships, and recall questions; feynman mode simplifies the material into plain language, highlights knowledge gaps, corrects them against the source, and devises a review plan; book mode processes an entire book chapter by chapter. Furthermore, DeepRead can compare 2–10 documents to uncover agreements, conflicts, and complementary evidence. There are two ways to utilize DeepRead: as an agent skill compatible with Codex and Claude, and as a plug-in for DeepSeek Harness. The plug-in includes a reading panel, PDF extraction, background jobs, live progress tracking, batch comparisons, cost estimates, and export options in Markdown, HTML, or XMind formats. To try it out, visit the GitHub repository: https://github.com/xiehuan123/dsh-deepread or download version v0.5.4 from the releases page at https://github.com/xiehuan123/dsh-deepread/releases/tag/v0.5.4. The repository features three authentic output examples rather than fabricated samples. DeepRead has been incorporated into several DeepSeek Harness and DSH community collections and its GitHub submission to Awesome Copilot has cleared all automated specifications, lint checks, installation tests, and version validations, awaiting maintainer approval. It is released under the MIT license. The project seeks input on a specific aspect: Is explicit claim-to-evidence tracing beneficial in your reading routine, or does it render the report overly cumbersome? The developers welcome issues and pull requests for further refinement. Suggested DEV tags: showdev, opensource, ai, productivity. Suggested Show HN title: Show HN: DeepRead – Evidence-first reading for books, articles, and PDFs.",
  "summary": "I recently built an open-source project called DeepRead to solve a problem I kept seeing with AI reading tools. Most tools can summarize a document quickly, but they often blur the line between the author's actual claims, the evidence provided by the source, and the model's own inferences. With long documents, source locations and the context around numerical claims can also disappear. DeepRead…",
  "key_points": [
    "DeepRead addresses AI reading tools' flaw in distinguishing original claims from AI deductions.",
    "Offers five reading modes: quick, deep, map, feynman, and book.",
    "Users can compare documents to find agreements, conflicts, and complementary evidence."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}