Show HN: HyperFlow – 基于 LangGraph 构建的自改进智能体框架
4 分•作者: lablnet•4 个月前
大家好，我是 Umer。我最近构建了一个名为 HyperFlow 的实验性框架，用于探索自我改进 AI 智能体的想法。
通常，当一个智能体未能完成任务时，我们开发者会介入手动调整提示词或修改代码逻辑。我想看看智能体是否可以自动化其自身的改进循环。
HyperFlow 基于 LangChain 和 LangGraph 构建，使用两个智能体：
* 一个 TaskAgent，用于解决领域问题。
* 一个 MetaAgent，充当改进者。
MetaAgent 观察 TaskAgent 的评估日志，重写底层的 Python 代码、工具和提示词文件，然后在隔离的沙盒（如 Docker）中测试新版本。经过几代迭代后，它会将获得最高分数的版本保存到档案中。
目前这还处于高度实验阶段，但其架构很大程度上受到了最近的 HyperAgents 论文（Meta Research，2026 年）的启发。
我很乐意听取您对该架构的反馈，您对自指智能体的看法，或者回答您可能有的任何问题！
文档：[https://hyperflow.lablnet.com/](https://hyperflow.lablnet.com/)
GitHub：[https://github.com/lablnet/HyperFlow](https://github.com/lablnet/HyperFlow)
查看原文
Hi HN, I am Umer. I recently built an experimental framework called HyperFlow to explore the idea of self-improving AI agents.<p>Usually, when an agent fails a task, we developers step in to manually tweak the prompt or adjust the code logic. I wanted to see if an agent could automate its own improvement loop.<p>Built on LangChain and LangGraph, HyperFlow uses two agents:
- A TaskAgent that solves the domain problem.
- A MetaAgent that acts as the improver.<p>The MetaAgent looks at the TaskAgent's evaluation logs, rewrites the underlying Python code, tools, and prompt files, and then tests the new version in an isolated sandbox (like Docker). Over several generations, it saves the versions that achieve the highest scores to an archive.<p>It is highly experimental right now, but the architecture is heavily inspired by the recent HyperAgents paper (Meta Research, 2026).<p>I would love to hear your feedback on the architecture, your thoughts on self-referential agents, or answer any questions you might have!<p>Documentation: <a href="https://hyperflow.lablnet.com/" rel="nofollow">https://hyperflow.lablnet.com/</a>
GitHub: <a href="https://github.com/lablnet/HyperFlow" rel="nofollow">https://github.com/lablnet/HyperFlow</a>