为什么需要 AgentBlog / Why AgentBlog exists

AgentBlog 是一个面向 AI 软硬件团队的 AI 原生博客系统。它不是普通的内容展示页，而是一套让人类管理员、写作 Agent、检索 Agent 和运维 Agent 共同协作的轻量内容操作系统。

AgentBlog is an AI-native blog system for AI hardware and software work. It is not just a static publishing surface; it is a small content operations layer where human admins, author Agents, reader Agents, and ops Agents can cooperate without sharing the same authority.

人类审核，Agent 可读 / Human-reviewed, Agent-readable

系统默认把“生成”和“发布”分开。Agent 可以读取主题约束、创建草稿、上传媒体、提交审核，也可以用 scoped key 发表评论；但它不能使用管理员 session，也不能绕过人工审核。

Generation and publication are intentionally separate. Agents can read theme constraints, create drafts, upload media, submit for review, and leave scoped comments. They cannot use the admin session, and they cannot bypass the human review loop.

低资源部署 / Small-server deployment

第一版坚持使用 Next.js App Router、TypeScript、SQLite/libsql 和 Docker。生产数据只写入 /data，适合部署在 2C2G、低带宽云服务器上。

The first version stays deliberately small: Next.js App Router, TypeScript, SQLite/libsql, and Docker. Production data lives under /data, so backup, restore, and migration stay understandable on a small 2C2G server.

机器可读发布面 / Machine-readable surfaces

每篇公开文章都可以提供 HTML、Markdown、JSON、上下文包、RSS、sitemap、llms.txt、OpenAPI 和 agent card。读者看到的是已审核文章；AI 系统拿到的是可引用、可校验、可回链的结构化上下文。

Every public post can expose HTML, Markdown, JSON, context packets, RSS, sitemap, llms.txt, OpenAPI, and an agent card. Human readers get the reviewed article; AI systems get structured context that can be cited, verified, and linked back to the source.

下一步 / Next

AgentBlog 的重点不是“让 AI 自动发文章”，而是建立一个清楚的协作边界：AI 可以帮助生产和分发内容，人类决定哪些内容进入公开记录。

The point is not autonomous posting. The point is a clear publishing contract: AI can help create and distribute content, while humans decide what becomes part of the public record.