https://zerocoding.online/ZeroCodingHugo 0.165.0 & FixIt v1.0.0-alpha.2-mtmrzvotzh-CNhi@zerocoding.online (ZeroCoding)hi@zerocoding.online (ZeroCoding)Thu, 03 Sep 2026 00:00:00 +0800https://zerocoding.online/posts/ai-agent-intro/Thu, 03 Sep 2026 00:00:00 +0800hi@zerocoding.online (ZeroCoding)https://zerocoding.online/posts/ai-agent-intro/AI<p>AI Agent 是当下大模型应用最热门的方向之一。本文从最基础的 ReAct 推理循环讲起，梳理单 Agent 的核心构成：模型、工具、记忆与规划。</p> <h2 class="heading-element" id="从-react-开始"><span>从 ReAct 开始</span> <a href="#%e4%bb%8e-react-%e5%bc%80%e5%a7%8b" class="heading-mark"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg></a> </h2><p>ReAct（Reasoning + Acting）把「思考」和「行动」交替进行，让模型在推理的同时调用外部工具，从而突破纯文本生成的局限。</p> <pre><code>思考 → 行动 → 观察 → 思考 → ... → 最终答案</code></pre><h2 class="heading-element" id="走向多智能体"><span>走向多智能体</span> <a href="#%e8%b5%b0%e5%90%91%e5%a4%9a%e6%99%ba%e8%83%bd%e4%bd%93" class="heading-mark"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg></a> </h2><p>当任务变复杂时，单一 Agent 会力不从心，于是有了多智能体协作：专家分工、流水线、辩论等模式。工程落地时，重点要处理好 Agent 间的消息传递、状态管理与容错。</p> <p>后续会结合具体框架展开实践。</p> https://zerocoding.online/posts/rag-practice/Thu, 20 Aug 2026 00:00:00 +0800hi@zerocoding.online (ZeroCoding)https://zerocoding.online/posts/rag-practice/AI<p>RAG（检索增强生成）是让大模型「有据可查」的关键手段。落地时，检索质量直接决定了生成效果的上限。</p> <h2 class="heading-element" id="选型三件套"><span>选型三件套</span> <a href="#%e9%80%89%e5%9e%8b%e4%b8%89%e4%bb%b6%e5%a5%97" class="heading-mark"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg></a> </h2><ol> <li><strong>Embedding 模型</strong>：中文场景优先选对中文友好的模型，注意向量维度与语种覆盖。</li> <li><strong>向量库</strong>：小规模用 pgvector 即可，规模上来再考虑 Milvus、Qdrant 等专用方案。</li> <li><strong>召回策略</strong>：稠密召回为主，稀疏（BM25）做补充，混合检索通常效果更稳。</li> </ol> <h2 class="heading-element" id="两个常被忽略的优化"><span>两个常被忽略的优化</span> <a href="#%e4%b8%a4%e4%b8%aa%e5%b8%b8%e8%a2%ab%e5%bf%bd%e7%95%a5%e7%9a%84%e4%bc%98%e5%8c%96" class="heading-mark"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg></a> </h2><ul> <li><strong>Chunk 切分</strong>：按语义边界切分，保留标题层级，比固定长度效果好很多。</li> <li><strong>重排序</strong>：用 Cross-Encoder 对召回结果二次排序，能明显提升 Top-K 的命中率。</li> </ul> https://zerocoding.online/posts/prompt-engineering/Mon, 10 Aug 2026 00:00:00 +0800hi@zerocoding.online (ZeroCoding)https://zerocoding.online/posts/prompt-engineering/AI<p>提示工程是把大模型用好的基本功。这里整理几个在真实业务里反复验证有效的技巧。</p> <h2 class="heading-element" id="三个核心原则"><span>三个核心原则</span> <a href="#%e4%b8%89%e4%b8%aa%e6%a0%b8%e5%bf%83%e5%8e%9f%e5%88%99" class="heading-mark"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg></a> </h2><ol> <li><strong>给足上下文</strong>：把背景、角色、约束写清楚，别让模型猜。</li> <li><strong>要求结构化输出</strong>：明确 JSON Schema 或 Markdown 格式，方便下游解析。</li> <li><strong>分步引导</strong>：复杂任务拆成多步，配合思维链（CoT）让模型「想清楚再说」。</li> </ol> <h2 class="heading-element" id="少样本示例"><span>少样本示例</span> <a href="#%e5%b0%91%e6%a0%b7%e6%9c%ac%e7%a4%ba%e4%be%8b" class="heading-mark"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg></a> </h2><p>给 1～3 个高质量示例，比写一堆抽象规则更有效。示例要覆盖边界情况，让模型学会「什么时候该怎么做」。</p> <p>这些技巧不花哨，但胜在稳定、可复现。</p> https://zerocoding.online/posts/idempotency-design/Thu, 30 Jul 2026 00:00:00 +0800hi@zerocoding.online (ZeroCoding)https://zerocoding.online/posts/idempotency-design/后端<p>幂等性是分布式系统的老问题，却是高并发场景下最容易踩坑的地方。核心诉求只有一个：<strong>同一个操作执行多次，结果和只执行一次一致</strong>。</p> <h2 class="heading-element" id="常见落地手段"><span>常见落地手段</span> <a href="#%e5%b8%b8%e8%a7%81%e8%90%bd%e5%9c%b0%e6%89%8b%e6%ae%b5" class="heading-mark"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg></a> </h2><ol> <li><strong>数据库唯一约束</strong>：最可靠，通过唯一索引天然去重，适合「创建」类操作。</li> <li><strong>状态机</strong>：把业务状态流转设计成幂等的，重复请求自然被挡在状态判断之外。</li> <li><strong>Token / 去重表</strong>：请求先拿 Token，消费一次即失效，适合「提交」类操作。</li> <li><strong>分布式锁</strong>：保证同一时刻只有一个请求在处理，但要处理好锁的粒度与超时。</li> </ol> <h2 class="heading-element" id="取舍"><span>取舍</span> <a href="#%e5%8f%96%e8%88%8d" class="heading-mark"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg></a> </h2><p>没有银弹。唯一约束最稳但局限性强，锁最灵活但引入复杂度。实际项目里往往是组合使用，按业务语义选择最合适的那一层。</p> https://zerocoding.online/posts/redis-ha/Sun, 12 Jul 2026 00:00:00 +0800hi@zerocoding.online (ZeroCoding)https://zerocoding.online/posts/redis-ha/后端<p>Redis 的高可用，最常用的两条路是哨兵（Sentinel）和集群（Cluster）。选型之前，先想清楚自己的真实需求。</p> <h2 class="heading-element" id="哨兵sentinel"><span>哨兵（Sentinel）</span> <a href="#%e5%93%a8%e5%85%b5sentinel" class="heading-mark"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg></a> </h2><p>主从 + 哨兵监控，主挂自动切换。<strong>适合数据量不大、但对可用性有要求的场景</strong>。优点是简单、兼容性好，缺点是数据集中在一台主节点上，受内存上限制约。</p> <h2 class="heading-element" id="集群cluster"><span>集群（Cluster）</span> <a href="#%e9%9b%86%e7%be%a4cluster" class="heading-mark"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg></a> </h2><p>数据分片到多个节点，天然支持水平扩展。<strong>适合数据量大、需要分片的场景</strong>。代价是运维复杂度上升，部分命令（如多 key 跨 slot 操作）受限。</p> <h2 class="heading-element" id="我的判断"><span>我的判断</span> <a href="#%e6%88%91%e7%9a%84%e5%88%a4%e6%96%ad" class="heading-mark"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg></a> </h2><ul> <li>单实例内存 &lt; 8G，优先哨兵，简单可靠。</li> <li>数据增长快、预期会分片，直接上 Cluster，避免中途迁移。</li> </ul> <p>迁移路径上，从哨兵到 Cluster 可以借助官方工具或双写方案平滑过渡。</p>