Show HN: Tiptap AI 工具包 – 现已探索服务端智能文档编辑
2 分•作者: philipisik•7 个月前
我们最近发布了 Tiptap AI 工具包 (<a href="https://tiptap.dev/docs/content-ai/capabilities/ai-toolkit/overview">https://tiptap.dev/docs/content-ai/capabilities/ai-toolkit/o...</a>)，它允许开发者将由 LLM 驱动的编辑功能嵌入到富文本编辑器中。
现在，我们正在探索下一步：让智能体在没有浏览器参与的情况下操作文档。
我们的想法是建立一个服务端系统，智能体可以在其中实时读取、写入和转换文档——本质上是将您的文档层视为一个可编程的数据存储，智能体可以直接与之交互。在内部，我们一直在开玩笑地称之为“实时智能体文档数据库”，因为它开始呈现出这样的形态。
我很好奇其他人会如何看待这个方向。
如果有一个数据库，其基本单元不是行/对象，而是具有语义结构的文档，并且智能体可以对其进行修改，那么它还能做或应该做什么？
有哪些显而易见的陷阱？
您希望在智能体自主性和文档完整性之间设置什么样的界限？
很想听听您将如何扩展或限制这样的系统。
查看原文
We recently shipped the Tiptap AI Toolkit (<a href="https://tiptap.dev/docs/content-ai/capabilities/ai-toolkit/overview">https://tiptap.dev/docs/content-ai/capabilities/ai-toolkit/o...</a>), which lets developers embed LLM-powered editing into rich text editors.<p>Now we’re exploring the next step: making agents operate on documents without a browser involved.<p>The idea is a server-side system where agents can read, write, and transform documents in real time — essentially treating your document layer as a programmable data store that agents can interact with directly. Internally we’ve been joking about the name “real-time agentic document database”, because that’s the shape it’s starting to take.<p>I’m curious how others would think about this direction.<p>If you had a database where the fundamental unit wasn’t rows/objects, but documents with semantic structure that agents can mutate, what else could or should it do?<p>Where are the obvious footguns?<p>What boundaries would you want between agent autonomy and document integrity?<p>Would love to hear how you’d expand or constrain a system like this.