docling-project/docling
Get your documents ready for gen AI
Python · ★ 67,344 · Fork 4,846 · 在 GitHub 查看
在手机上浏览源码、订阅每日 AI 趋势日报与项目架构深度解读:获取 CopoHub 客户端(iOS & HarmonyOS)→
GitHub Trending 上榜记录
docling-project/docling 已在 GitHub Trending 每日榜单上榜 2 天,最高排名第 8,首次上榜 2026-09-19,最近一次上榜 2026-09-20。
| 日期 | 排名 | Star | 当日增长 |
|---|---|---|---|
| 2026-09-20 | #8 | 67,344 | +129 |
| 2026-09-19 | #8 | 67,000 | +94 |
同期上榜的项目
- affaan-m/ECC – The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
- BuilderIO/agent-native – A framework for building agentic apps
- cloudflare/security-audit-skill – A coding-agent skill for multi-phase security audits with independently verified, machine-readable findings
- trycua/cua – Scale computer-use 2.0 with open-source drivers, cross-OS sandboxes, and benchmarks for training, evaluation, and data generation.
- anthropics/financial-services
- paperless-ngx/paperless-ngx – A community-supported supercharged document management system: scan, index and archive all your documents
Docling 项目深度剖析与商业评估报告
Docling 项目深度剖析与商业评估报告
1. 核心定位与商业落地价值
Docling 由 IBM Research Zurich 的 AI for Knowledge 团队发起,现已捐赠至 LF AI & Data 基金会,定位为"面向生成式 AI 的文档预处理基础设施"。其根本痛点在于:RAG 与 Agent 系统的检索质量上限往往由文档解析保真度决定。传统 PyPDF2/pdfplumber 只能抽取裸文本,丢失阅读顺序、表格结构、公式与图表语义,导致下游向量检索召回噪声极大。Docling 通过统一的 DoclingDocument 中间表示(IR),将 PDF/DOCX/PPTX/XLSX/HTML/EPUB/音频/视频等异构输入规整为带层级与语义标注的结构化对象。
企业级落地场景: 金融与合规:XBRL 财报、USPTO 专利、JATS 学术论文的 schema 原生支持,可直接对接监管报送与尽调流水线。 制药/法律/国防:本地执行与 air-gapped 部署能力,满足数据不出域的硬性合规要求。 Agentic RAG:原生集成 LangChain、LlamaIndex、Haystack、CrewAI,并通过 MCP Server 暴露为 Agent 工具。
竞争差异化与护城河: 1. 格式广度:竞品(Unstructured.io、LlamaParse、Marker)多聚焦 PDF/Markdown,Docling 覆盖音视频、ODF、Apple Pages、邮件等长尾格式。 2. IBM 背书 + 开源治理:MIT 许可 + LF AI & Data 中立托管,规避了 LlamaParse 等 SaaS 方案的供应商锁定风险。 3. 统一 IR 的生态位:DoclingDocument 类似"文档领域的 ONNX",可同时导出 Markdown/DocTags/JSON/DocLang,具备成为事实标准的潜力。
2. 核心架构设计与技术深度
Docling 采用分层管道(Pipeline)+ 插件式后端架构,核心链路如下:
Input (URL/Path/Bytes) → Format Detection & Backend Router → Per-Format Parser (PDF Backend / MS Office / HTML / ASR) → Layout Analysis (Layout Model) → Reading Order Reconstruction → Table Structure Recognition (TableFormer) → OCR (EasyOCR/Tesseract/RapidOCR) — 按需触发 → DoclingDocument (Unified IR) → Exporter (Markdown / HTML / JSON / DocTags)
在 README 中展示上榜徽章
[](https://github.fq6825.top/repo/docling-project/docling)