Show HN: Agentlearn – AI智能体基础交互式课程
1 分•作者: init0•6 个月前
<a href="https://agentlearn.dev" rel="nofollow">https://agentlearn.dev</a>
我注意到大多数 AI 智能体教程都侧重于框架(LangChain、CrewAI)而非基础知识,因此我构建了 agentlearn。这导致开发者只会复制粘贴代码,但在出现问题时却束手无策。
这是一个免费的互动课程,涵盖以下内容:
* 智能体循环 - 为什么循环至关重要(思考 vs. 行动 vs. 观察)
* 上下文工程 - “提示工程”背后的真正技能
* 工具 & 函数调用 - 将文本生成与实际行动连接起来
* 记忆系统 - 短期 vs. 长期,向量数据库
* 协议 - MCP、A2A 以及新兴标准
* 生产模式 - 错误处理、成本优化、可观测性
每个概念都有可运行的代码沙盒,你可以逐步操作。设计上故意采用“手绘”风格,使其看起来比典型的技术文档更亲切。
技术栈:原生 JS + Vite,无框架。
为什么不使用框架?因为理解基础知识意味着理解框架抽象了什么。一旦你掌握了核心循环,你就可以使用任何框架——或者构建你自己的框架。
欢迎反馈!特别感兴趣的是缺少哪些主题。
查看原文
<a href="https://agentlearn.dev" rel="nofollow">https://agentlearn.dev</a><p>I built agentlearn after noticing that most AI agent tutorials focus on frameworks (LangChain, CrewAI) rather than fundamentals. The result is developers who can copy-paste code but struggle when things break.<p>This is a free, interactive course covering:<p>The Agent Loop - Why loops matter (think vs. act vs. observe)
Context Engineering - The real skill behind "prompt engineering"
Tools & Function Calling - Bridging text generation to real actions
Memory Systems - Short-term vs. long-term, vector DBs
Protocols - MCP, A2A, and the emerging standards
Production Patterns - Error handling, cost optimization, observability
Each concept has runnable code sandboxes you can step through. The design is intentionally "hand-drawn" to feel less intimidating than typical technical docs.<p>Tech stack: Vanilla JS + Vite, no framework.<p>Why no framework? Because understanding fundamentals means understanding what frameworks abstract away. Once you get the core loop, you can use any framework—or build your own.<p>Feedback welcome! Especially interested in what topics are missing.