Help：How to build a knownledge base? I should create with a single file or split the file into some smale file part?

[simonjhy]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:)
• Please do not modify this template :) and fill in all the required fields.

1. Is this request related to a challenge you're experiencing? Tell me about your story.

If my knowledge base corresponds to a single chip datasheet, should I split the datasheet into multiple files when creating the knowledge base, or keep it as a single file?
Because if the datasheet is updated later, how should I perform incremental updates? Or does having only one file mean that I can only delete the existing file to update the knowledge base?
Alternatively, do I need to split the datasheet into multiple files when creating the knowledge base so that I can support incremental updates later, without having to re-vectorize the entire knowledge base?

2. Additional context or comments

No response

[dosubot[bot]]

For Dify knowledge bases, it's best to split a large datasheet into multiple smaller files or logical sections (for example, by chapter, table, or topic). This approach makes incremental updates much easier: if part of the datasheet changes, you can update or replace just the relevant file(s) or segment(s), and only those parts will be re-indexed and re-vectorized—not the entire knowledge base. Dify's API and backend are designed to support creating, updating, and deleting individual documents and segments, so you don't need to delete and re-upload everything for small changes. This modular structure is more efficient and aligns with how Dify manages vector updates and indexing tasks see documentation and backend implementation.

If you use a single large file, any update will require reprocessing the whole file. Splitting into smaller files or logical parts is the recommended way to support efficient incremental updates in Dify.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[simonjhy]

那对增量的更新，我是否再可以新建一个知识库来存放新增或者修改部分的内容，对他们进行索引和向量化？但是大模型会区分他们的先后顺序吗？能通过更新的时间来确定出来先后顺序吗？

[dosubot[bot]]

可以，你可以新建一个知识库（Dataset）来存放新增或修改的内容，并对其进行索引和向量化。Dify 支持在一次检索中同时查询多个知识库，结果会合并并根据相似度或重排序模型的得分进行排序，这样你可以实现增量更新，无需每次都重建整个知识库 参考。

但需要注意，大模型（LLM）和 Dify 检索流程本身不会自动区分内容的“先后顺序”或“更新时间”，检索结果默认只按相似度得分排序，而不是按更新时间或创建时间排序。虽然每条文档和知识库都有内置的 created_at 和 updated_at 字段（可用于元数据过滤），但这些时间信息不会自动影响检索排序 参考。

如果你有“最新内容优先”或“按更新时间排序”的需求，可以在检索时通过元数据过滤（前提是所有知识库都有相同的元数据字段），或者在检索结果返回后自行做二次排序和处理 参考。

最佳实践是将变更内容直接更新到同一个知识库，或者新建知识库并在检索时一起引用。如果需要按时间优先级处理，需要你自己实现相关逻辑。需要具体实现建议可以补充说明你的场景。

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}