我为何用 15KB 的 httpx 替换了 LangChain
📄 中文摘要
文章作者在采用 LangChain 六个月后，决定将其完全替换为仅使用 httpx 库、约 500 行 Python 代码的自定义解决方案。此举带来了显著的性能提升，处理速度加快了三倍，从 420 毫秒缩短至 165 毫秒。同时，测试覆盖率从 61% 大幅提升至 94%，彻底消除了依赖问题，并且总代码量仅为 15KB。作者指出，LangChain 的主要问题在于过度抽象，将每个 API 调用都封装在多层结构中，导致代码复杂性增加和性能下降。通过直接使用 httpx，开发者能够更精细地控制底层 HTTP 请求，从而优化效率和可维护性。这一实践案例揭示了在特定场景下，精简的自定义实现可能比功能丰富的框架更具优势，尤其是在对性能、代码体积和依赖管理有严格要求的项目中。
📄 English Summary
Why I Replaced LangChain with 15KB of httpx
The author, after six months of using LangChain, opted to replace the entire framework with a custom solution built using only the httpx library, comprising approximately 500 lines of Python code. This transition yielded significant improvements: a threefold increase in speed, reducing processing time from 420ms to 165ms. Furthermore, test coverage dramatically improved from 61% to 94%, dependency issues were completely eliminated, and the total code size was a mere 15KB. The core issue identified with LangChain was its 'abstraction overload,' where every API call was wrapped in multiple layers, leading to increased complexity and performance degradation. By directly utilizing httpx, the developer gained finer control over underlying HTTP requests, optimizing both efficiency and maintainability. This case study highlights that in specific contexts, a lean, custom implementation can outperform feature-rich frameworks, particularly for projects with stringent requirements for performance, code footprint, and dependency management.