Show HN: Docgen – C++ AI 命令行工具，利用本地 LLM 解决文档编写难题
1 分•作者: alonsovm•5 个月前
Hi HN，
我是一个独立开发者，厌倦了“文档地狱”，要么花几个小时写文档，结果文档很快就过时了，要么干脆没有文档。我想要一个工具，将文档生成视为一个标准的构建步骤，所以我开发了Docgen。
Docgen是一个用C++编写的轻量级AI CLI工具，它将文档视为代码进行处理。它位于你的代码库中（通过一个.docgen文件夹和一个Docfile），并在你的源代码旁边生成Markdown文件。
以下是一些关于其工作原理的技术细节：
- 本地优先 & 私有：它默认使用本地的Ollama，这样你的专有代码永远不会离开你的机器（尽管如果你愿意，它也支持像OpenAI/Gemini这样的云API）。
- 智能增量构建：它使用内容哈希。当你运行docgen update时，它只重新生成实际更改过的文件的文档，从而节省大量的API积分和计算时间。
- 上下文感知（RAG）：它会自动分析#include依赖关系，为LLM提供正确的上下文，而不是盲目地向它提供单个文件。
- 零依赖：编译成一个单一的静态二进制文件。只需下载并运行即可。
- “自动”模式：这是我最喜欢的部分。如果你运行“docgen auto”，它会充当一个文件监视器，并内置了防抖动功能（在你停止输入/保存后等待几秒钟）。它会在后台默默地更新你的Markdown文档，让你保持流畅的开发状态。
我目前专注于改进RAG上下文处理。
你可以在这里查看：[https://github.com/alonsovm44/docgen](https://github.com/alonsovm44/docgen)
我很乐意听取你的想法，对架构的批评，或者你认为我应该处理的任何边缘情况！
查看原文
Hi HN,<p>I’m a solo dev who got tired of the "documentation hell", either spending hours writing docs that immediately become outdated, or having no docs at all. I wanted a tool that treats documentation generation as a standard build step, so I built Docgen.<p>Docgen is a lightweight AI CLI tool written in C++ that automates docs-as-code. It sits in your repo (via a .docgen folder and a Docfile) and generates Markdown files next to your source.<p>A few technical details on how it works under the hood:<p>- Local-First & Private: It defaults to using Ollama locally so your proprietary code never leaves your machine (though it supports cloud APIs like OpenAI/Gemini if you prefer).<p>- Smart Incremental Builds: It uses content hashing. When you run docgen update, it only regenerates docs for files that actually changed, saving massive amounts of API credits and compute time.<p>- Context-Aware (RAG): It automatically analyzes #include dependencies to give the LLM the right context, rather than just blind-feeding it a single file.<p>- Zero Dependencies: Compiled as a single static binary. Just download and run.<p>- The "Auto" Mode: This is my favorite part. If you run "docgen auto", it acts as a file watcher with a built-in debounce (waits a few seconds after you stop typing/saving). It quietly updates your Markdown docs in the background while you stay in your flow state.<p>I’m currently focused on improving the RAG context handling.<p>You can check it out here: <a href="https://github.com/alonsovm44/docgen" rel="nofollow">https://github.com/alonsovm44/docgen</a><p>I'd love to hear your thoughts, critique on the architecture, or any edge cases you think I should handle!