To be honest, it is easier than I expected to build an agent. Tool calling is really straightforward, which shows how powerful LLMs have become through training! It used to be not that easy for LLMs to generate Structured Outputs.
Here is a minimal CLI agent built in TypeScript following the ReAct (Reasoning + Acting) pattern.
查看全文Kubernetes是一种实现分布式架构的框架，本文将展示如何在单台主机上运行分布式架构的项目用于调试，并最终打包便于部署。
查看全文Harness Engineering是能让AI agent的产出更稳定的工程方法。按照Anthropic的文章，通过分设三角的方式，避免agent因为context污染产生幻觉。
查看全文本文将使用OpenWRT上的natmap，设置防火墙规则开放和转发端口，然后以qBittorrent为例，将通告端口改为公网端口。
查看全文In the study of SICP (Structure and Interpretation of Computer Programs), there is a homework in Lecture 1B, MIT 6.001 (1986).
查看全文Android上使用字体使用OTC和可变字体时需要修改fonts.xml，而Noto CJK需要删除一些cmap条目。
查看全文You can use variable font for some writing system after Android Oreo.
查看全文Hide top divider while scrolling to the top, hide bottom divider while scrolling to the end.
查看全文