![\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030530-69a25b7a35714.png\")
<\/p>\n\u5e74\u521d\u706b\u7206\u73b0\u5728\u70ed\u5ea6\u4f9d\u65e7\u7684OpenClaw\u4e3aAI Agent\u5e26\u6765\u4e86\u65b0\u7684\u60f3\u8c61\u3002\u5982\u679c\u8bf42025\u662fAI Agent\u667a\u80fd\u4f53\u5143\u5e74，\u60f3\u5fc52026\u5c06\u4f1a\u662fAI Agent\u771f\u6b63\u5546\u7528\u5316\u7684\u5f00\u7aef，\u800cAI Agent\u5546\u7528\u5316\u7684\u524d\u63d0\u662f\u5404\u884c\u5404\u4e1a\u5f00\u59cb\u843d\u5730\u5b9e\u9645\u5546\u4e1a\u9886\u57df\u7684AI Agent\u3002<\/p>\n
\u4f5c\u4e3a\u4e00\u540d\u7a0b\u5e8f\u5458\/\u5de5\u7a0b\u5e08，\u601d\u8003\u7684\u66f4\u591a\u7684\u53ef\u80fd\u662f\u5982\u4f55\u5c06AI Agent\u843d\u5730\u5b9e\u73b0\u7684\u5c42\u9762\u3002\u5de5\u7a0b\u6846\u67b6\u5f80\u5f80\u662f\u5de5\u7a0b\u5e94\u7528\u5b9e\u73b0\u7684\u57fa\u77f3，\u6846\u67b6\u9009\u578b\u4e5f\u662f\u67b6\u6784\u5e08\u4eec\u8bbe\u8ba1\u4e0e\u5b9e\u73b0\u4e00\u4e2aAI Agent\u667a\u80fd\u4f53\u7684\u9996\u8981\u5de5\u4f5c\u3002\u7531\u4e8e\u7bc7\u5e45\u8fc7\u957f，\u672c\u6587\u5c06\u5206\u6587\u4e24\u4e2a\u90e8\u5206\u6765\u5199：\u4e00\u7406\u8bba\u7bc7 \u4e0e \u4e8c\u5b9e\u8df5\u7bc7\u3002<\/p>\n
Agent = Reasoning + Acting<\/p>\n
<\/p>\n
1.1 AI Agent \u6846\u67b6\u57fa\u7840\u7406\u8bba<\/p>\n
AI \u667a\u80fd\u4f53\u662f\u4f7f\u7528 AI \u6765\u5b9e\u73b0\u76ee\u6807\u5e76\u4ee3\u8868\u7528\u6237\u5b8c\u6210\u4efb\u52a1\u7684\u8f6f\u4ef6\u7cfb\u7edf\u3002\u5176\u8868\u73b0\u51fa\u4e86\u63a8\u7406\u3001\u89c4\u5212\u548c\u8bb0\u5fc6\u80fd\u529b，\u5e76\u4e14\u5177\u6709\u4e00\u5b9a\u7684\u81ea\u4e3b\u6027，\u80fd\u591f\u81ea\u4e3b\u5b66\u4e60\u3001\u9002\u5e94\u548c\u505a\u51fa\u51b3\u5b9a\u3002<\/p>\n
Google Cloud<\/p>\n
\u5173\u4e8eAI Agent\u8fd9\u91cc\u5f15\u7528\u4e86Google Cloud\u7684\u5b9a\u4e49，\u8a00\u7b80\u610f\u8d45\u3002<\/p>\n
1）ReAct \u6a21\u5f0f<\/p>\n
\u5728\u5f53\u524dAI Agent\u7406\u8bba\u4e2d，\u6700\u5177\u6709\u57fa\u7840\u6027\u4e0e\u4ee3\u8868\u6027\u7684\u8981\u6570ReAct\u6a21\u5f0f，\u5b83\u662fYao\u7b49\u4eba\u4e8e2022\u5e74\u5728\u300aReAct: Synergizing Reasoning and Acting in Language Models\u300b\u8bba\u6587\u4e2d\u63d0\u51fa，\u6838\u5fc3\u601d\u60f3\u5c06\u63a8\u7406（Reasoning）\u548c\u884c\u52a8（Acting）\u76f8\u7ed3\u5408\u3002CoT\u63d0\u5347\u7684\u662fLLM\u7684\u63a8\u7406\u80fd\u529b，\u4f46\u5b83\u7684\u7f3a\u70b9\u5728\u4e8e\u7f3a\u5c11\u4e0e\u5916\u90e8\u4e16\u754c\u7684\u4ea4\u4e92\u4ece\u800c\u7f3a\u5c11\u5916\u90e8\u53cd\u9988\u6765\u62d3\u5c55\u81ea\u8eab\u7684\u77e5\u8bc6\u7a7a\u95f4，ReAct\u5f25\u8865\u4e86\u8fd9\u4e00\u7f3a\u9677\u3002<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030530-69a25b7ae3d3c.png\")
<\/p>\n
ReAct\u667a\u80fd\u4f53\u7684\u8fd0\u4f5c\u57fa\u4e8e\u4e00\u4e2a\u5faa\u73af\u8fc7\u7a0b（\u4e0d\u65ad\u8fed\u4ee3\u66f4\u65b0），\u5305\u62ec\u4ee5\u4e0b\u4e09\u4e2a\u6b65\u9aa4：<\/p>\n
2）Plan-and-Execute \u6a21\u5f0f<\/p>\n
2023\u5e745\u6708，Langchain\u56e2\u961f\u57fa\u4e8eLei Wang\u7b49\u53d1\u8868\u7684\u300aPlan-and-Solve Prompting\u300b\u8bba\u6587\u548c\u5f00\u6e90\u7684BabyAGI Agent\u9879\u76ee\u7684\u5de5\u4f5c，\u63d0\u51fa\u4e86Plan-and-Execute \u6a21\u5f0f\u3002<\/p>\n
Plan-and_Execute\u6a21\u5f0f\u5f3a\u8c03\u5148\u5236\u5b9a\u591a\u6b65\u8ba1\u5212，\u518d\u9010\u6b65\u6267\u884c，\u5c5e\u4e8e\u7ed3\u6784\u5316\u5de5\u4f5c\u6d41\u7a0b（Planing -> Task1 -> Task2 -> Task3 -> Summary），\u6bd4\u8f83\u9002\u5408\u590d\u6742\u4e14\u4efb\u52a1\u5173\u7cfb\u4ee5\u6765\u660e\u786e\u7684\u957f\u671f\u4efb\u52a1\u3002\u7f3a\u70b9\u5219\u662f\u503e\u5411\u4e8eworkflow，\u7f3a\u4e4f\u52a8\u6001\u8c03\u6574\u80fd\u529b\u3002<\/p>\n
3）Reflection \u6a21\u5f0f<\/p>\n
\u6700\u65e9\u7cfb\u7edf\u6027\u7684\u63d0\u51fa\u5728Agent\u4e2d\u5f15\u5165\u53cd\u601d\u6982\u5ff5\u7684\u662fNoah Shinn\u3001Shunyu Yao（\u5bf9，\u8fd8\u662fReAct\u4f5c\u8005）\u7b49\u7684\u300aReflexion: Language Agents with Verbal Reinforcement Learning\u300b\u8bba\u6587，\u8bba\u6587\u91cc\u63d0\u51fa\u4e86Reflexion\u6846\u67b6，\u901a\u8fc7\u8bed\u8a00\u53cd\u9988\u800c\u975e\u6743\u91cd\u66f4\u65b0\u6765\u5f3a\u5316\u8bed\u8a00Agent\u3002Agent\u901a\u8fc7\u5bf9\u4efb\u52a1\u53cd\u9988\u4fe1\u53f7\u8fdb\u884c \u53e3\u5934\u53cd\u601d，\u7136\u540e\u5728\u60c5\u8282\u8bb0\u5fc6\u7f13\u51b2\u533a\u4e2d\u7ef4\u62a4\u81ea\u5df1\u7684\u53cd\u601d\u6587\u672c，\u4ee5\u5728\u540e\u7eed\u8bd5\u9a8c\u4e2d\u505a\u51fa\u66f4\u597d\u7684\u51b3\u7b56\u3002<\/p>\n
\u53e6\u5916，Aman Madaan\u7b49\u53d7\u4eba\u7c7b\u6539\u8fdb\u6587\u672c\u65b9\u5f0f\u7684\u542f\u53d1，\u5728\u300aSelf-Refine: Iterative Refinement with Self-Feedback\u300b\u8bba\u6587\u63d0\u51fa\u4e86\u4e00\u79cd\u540dSelf-Refine\u7684\u65b9\u6cd5，\u901a\u8fc7\u8fed\u4ee3\u53cd\u9988\u548c\u6539\u8fdb\u6765\u63d0\u5347 LLM \u7684\u521d\u59cb\u8f93\u51fa：\u5148\u8ba9LLM\u8f93\u51fa，\u7136\u540e\u518d\u6839\u636e\u8f93\u51fa\u63d0\u4f9b\u53cd\u9988，\u4e0d\u65ad\u8fed\u4ee3\u3002\u5728\u6240\u6709\u8bc4\u4f30\u7684\u4efb\u52a1\u4e2dSelf-Refine\u65b9\u6cd5\u53ef\u4ee5\u4f7f\u5f97\u4efb\u52a1\u6027\u80fd\u5e73\u5747\u63d0\u5347\u7ea6 20%\u3002<\/p>\n
\u6e05\u534e\u5927\u5b66\u4e0e\u5fae\u8f6f\u8054\u5408\u53d1\u5e03\u7684\u300aCRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing\u300b\u8bba\u6587\u5219\u7ed3\u5408\u5916\u90e8\u5de5\u5177（\u5982\u641c\u7d22\u5f15\u64ce\u3001\u4ee3\u7801\u6267\u884c\u5668）\u9a8c\u8bc1\u8f93\u51fa，\u518d\u57fa\u4e8e\u9a8c\u8bc1\u7ed3\u679c\u81ea\u6211\u4fee\u3002<\/p>\n
\u8fd9\u4e9b\u91cc\u7a0b\u7891\u8bba\u6587\u90fd\u662fReflection\u6a21\u5f0f\u7684\u7406\u8bba\u57fa\u7840，\u5f53\u524d\u4e3b\u6d41\u7684Agent\u6846\u67b6\u867d\u7136\u6709\u5404\u79cd\u6f14\u7ece\u4e0e\u53d8\u5f62，\u4e5f\u90fd\u662f\u5728ReAct\u63d0\u51fa\u4e4b\u540e\u53d1\u5c55\u51fa\u6765\u7684\u6269\u5c55\u548c\u8865\u5145，Agent\u6838\u5fc3\u5b9e\u8df5\u4f9d\u65e7\u79bb\u4e0d\u5f00ReAct：\u5c06\u63a8\u7406\u4e0e\u6267\u884c\u7ed3\u5408\u8d77\u6765\u3002<\/p>\n
1.2 \u4e3b\u6d41 AI Agent \u6846\u67b6\u5bf9\u6bd4<\/p>\n
\u5f53\u524d\u4e3b\u6d41Agent\u6846\u67b6\u4e3b\u8981\u5305\u542b\u4ee5\u4e0b\u51e0\u79cd：<\/p>\n
\u9009\u62e9\u5efa\u8bae<\/p>\n
\u53e6\u5916，\u968f\u7740Anthropic\u516c\u53f8\u7684Claude Cowork AI\u901a\u7528Agent\u5174\u8d77，\u4e00\u4e9b\u57fa\u4e8e\u901a\u7528Code Agent SDK\u7684\u5957\u58f3Agent\u4e5f\u5f00\u59cb\u6d41\u884c（\u5982\u516c\u53f8CodeBuddy\u56e2\u961f\u57fa\u4e8eCodeBuddy Agent SDK\u7b49\u751f\u6210\u7684WorkBuddy\u5e94\u7528），\u8fd9\u4e9bAgent\u7684\u521b\u65b0\u4e4b\u5904\u5728\u4e8e\u53ef\u4ee5\u9488\u5bf9\u5404\u7c7b\u7528\u6237\u573a\u666f\u63d0\u4f9b\u66f4\u597d\u7684\u4ea4\u4e92\u8bbe\u8ba1\u4e0e\u5de5\u4f5c\u6d41\u573a\u666f\u89e3\u51b3\u65b9\u6848\u3002<\/p>\n
1.3 AI Agent \u6846\u67b6\u6838\u5fc3<\/p>\n
\u5728Agent\u5e94\u7528\u53d1\u5c55\u4e0e\u5b9e\u8df5\u8fc7\u7a0b\u4e2d，\u6709\u4e00\u5bb6\u516c\u53f8\u7684\u4e00\u6b3eAgent\u5e94\u7528\u4e0d\u5f97\u4e0d\u63d0，\u90a3\u5c31\u662fAI\u521d\u521b\u516c\u53f8Monica\u53d1\u5e03\u7684Agent C\u7aef\u4ea7\u54c1：Manus，\u5b83\u7684\u7206\u706b\u8ba9\u8ba9Agent\u4ea7\u54c1\u8fdb\u5165\u5927\u4f17\u89c6\u91ce\u3002<\/p>\n
"Actually, Manus doesn\u2019t use MCP"\u8fd8\u6709\u540e\u534a\u53e5：\u201cinspired by CodeAct\u201d\u3002CodeAct\u662f\u4e00\u4e2aAgent\u8bbe\u8ba1\u67b6\u6784，\u6765\u81ea\u4e8eUIUC\u7684\u738b\u661f\u5c27\u535a\u58eb\u57282024\u5e74\u521d\u53d1\u8868\u7684\u4e00\u7bc7\u8bba\u6587\u300aExecutable Code Actions Elicit Better LLM Agents\u300b，\u5b83\u63d0\u51fa\u901a\u8fc7\u751f\u6210\u53ef\u6267\u884c\u7684Python\u4ee3\u7801\u6765\u7edf\u4e00LLM Agent\u7684\u884c\u52a8\u7a7a\u95f4，\u4e5f\u5c31\u662fActing\u4e0d\u4ec5\u53ef\u4ee5\u6709Fucntion Call\u548cMCP，\u8fd8\u53ef\u4ee5\u6267\u884c\u4ee3\u7801\u5b8c\u6210\u4efb\u52a1，\u800c\u4e14\u6548\u679c\u66f4\u597d\u30022025\u5e7411\u6708，Anthropic\u5b98\u65b9\u535a\u5ba2\u66f4\u65b0\u6280\u672f\u6587\u7ae0\u300aCode execution with MCP: Building more efficient agents\u300b，\u63d0\u51fa\u5c06 MCP \u670d\u52a1\u5668\u4f5c\u4e3a\u4ee3\u7801 API（\u800c\u975e\u76f4\u63a5\u7684\u5de5\u5177\u8c03\u7528）\u6765\u63d0\u4f9b，\u7136\u540e，Agent\u5c31\u53ef\u4ee5\u7f16\u5199\u4ee3\u7801\u6765\u4e0e MCP \u670d\u52a1\u5668\u4ea4\u4e92，\u8fd9\u6837Agent\u53ef\u4ee5\u6309\u9700\u52a0\u8f7d，\u66f4\u9ad8\u6548\u5730\u5229\u7528\u4e0a\u4e0b\u6587\u3002CodeAct\u8fd9\u4e2a\u89c2\u70b9\u4e0e\u672c\u6587\u5f00\u5934\u7684Shunyu Yao\u7684"\u4eba\u7c7b\u6700\u91cd\u8981\u7684 affordance \u662f\u624b，\u800c AI \u6700\u91cd\u8981\u7684 affordance \u53ef\u80fd\u662f\u4ee3\u7801"\u4e5f\u4e0d\u8c0b\u800c\u5408\u3002<\/p>\n
\u4eceManus\u7684\u6545\u4e8b，\u6211\u4eec\u53ef\u4ee5\u5f97\u51fa\u5f53\u524d\u5173\u4e8eAgent\u5de5\u7a0b\u7684\u4e24\u5927\u4e1a\u5185\u5171\u8bc6：<\/p>\n
\u867d\u7136\u5f53\u524dAgent\u6846\u67b6\u53d1\u5c55\u4eceReAct\u6a21\u5f0f\u9010\u6e10\u878d\u5408CodeAct\u6a21\u5f0f，\u4f46\u662fAgent\u6846\u67b6\u672c\u8d28\u7684\u63a8\u7406\u4e0e\u6267\u884c\u529f\u80fd\u5e76\u6ca1\u6709\u53d8\u5316\u3002<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030531-69a25b7b1d1ee.png\")
<\/p>\n
\u5728\u5de5\u7a0b\u5c42\u9762\u6765\u8bf4，\u63a8\u7406\u672c\u8d28\u5c31\u662fLLM Call，\u6267\u884c\u672c\u8d28\u5219\u662fTools Call（\u4ee3\u7801\u53ef\u8ba4\u4e3a\u662fTools\u7684\u4e00\u79cd），\u800c\u8fde\u63a5\u8fd9\u4e8c\u8005\u7684\u4e0a\u4e0b\u6587\u5de5\u7a0b（Context Engineering）\u5219\u662fAgent\u6846\u67b6\u7684\u6838\u5fc3\u3002<\/p>\n
1.3.1 Agent \u6846\u67b6\u4e09\u5927\u90e8\u5206<\/p>\n
\u4e0b\u9762\u6211\u4eec\u6765\u62c6\u89e3\u4e00\u4e0bAgent\u6846\u67b6\u5728\u5de5\u7a0b\u4e0a\u7684\u89e3\u6cd5，\u4e3b\u8981\u5305\u62ec\u4e09\u5927\u90e8\u5206<\/p>\n
\u4ee5\u4e0aAgent\u4e09\u5927\u90e8\u5206\u4e2d，\u7b2c1\u90e8\u5206\u7684LLM Call\u57fa\u672c\u4e0a\u6ca1\u6709\u5565\u5de5\u7a0b\u53d8\u91cf，\u8fd9\u5757\u7684\u5de5\u7a0b\u5b9e\u8df5\u4e2dLiteLLM\u5e93（\u5b83\u662f\u4e00\u4e2aPython \u5e93，\u65e8\u5728\u7b80\u5316\u591a\u79cd\u5927\u578b\u8bed\u8a00\u6a21\u578b（LLM）API \u7684\u96c6\u6210）\u5df2\u7ecf\u662f\u4f7c\u4f7c\u8005\u4e86；\u7b2c2\u90e8\u5206\u7684Tools\u4f7f\u7528，\u5305\u542b\u7684\u5de5\u5177\u5217\u8868\u8303\u56f4\u4e5f\u6709\u4e1a\u5185\u6700\u4f73\u5b9e\u8df5，\u5177\u4f53\u53d6\u51b3\u4e8eAgent\u4f7f\u7528\u7684\u4e1a\u52a1\u573a\u666f\u3002<\/p>\n
\u5269\u4e0b\u7684\u6700\u5927\u7684\u4e00\u4e2a\u53d8\u91cf\u662f\u7b2c3\u90e8\u5206\u7684\u4e0a\u4e0b\u6587\u5de5\u7a0b，\u8fd9\u4e5f\u662fAgent\u6846\u67b6\u667a\u80fd\u7684\u6838\u5fc3\u6240\u5728\u3002OpenClaw\u7684\u7206\u706b\u51fa\u5708\u9664\u4e86\u5c06\u624b\u673a\u5373\u65f6\u5e94\u7528\u4e0eAgent\u7ed3\u5408\u8fd9\u70b9\u4e4b\u5916，\u5b83\u7684\u4e0a\u4e0b\u6587\u5de5\u7a0b\u7684\u7ba1\u7406\u4e5f\u975e\u5e38\u6709\u521b\u65b0\u3002<\/p>\n
\u8fd1\u65e5，Shunyu Yao\u56e2\u961f\u5728\u817e\u8baf\u6df7\u5143\u5b98\u7f51\u53d1\u8868\u4e86\u4e00\u7bc7\u540d\u4e3a\u300a\u4ece Context \u5b66\u4e60，\u8fdc\u6bd4\u6211\u4eec\u60f3\u8c61\u7684\u8981\u96be\u300b\u7684\u6587\u7ae0，\u63d0\u51fa\u201c\u6a21\u578b\u60f3\u8981\u8fc8\u5411\u9ad8\u4ef7\u503c\u5e94\u7528，\u6838\u5fc3\u74f6\u9888\u5c31\u5728\u4e8e\u80fd\u5426\u7528\u597d Context\u3002\u201d\u7684\u89c2\u70b9\u3002\u6b63\u5982\u6587\u4e2d\u6240\u8a00"\u5728\u4e0d\u63d0\u4f9b\u4efb\u4f55 Context \u7684\u60c5\u51b5\u4e0b，\u6700\u5148\u8fdb\u7684\u6a21\u578b GPT-5.1 (High) \u4ec5\u80fd\u89e3\u51b3\u4e0d\u5230 1% \u7684\u4efb\u52a1"\u3002\u5982\u679c\u8bf4Agent\u5e94\u7528\u4e2d，\u73b0\u5728\u54ea\u91cc\u8fd8\u662f\u4f4e\u5782\u7684\u679c\u5b9e\u53ef\u4ee5\u6458\u53d6，\u90a3\u53ef\u80fd\u662f\u4e0a\u4e0b\u6587\u5de5\u7a0b（Context Engineering）\u3002<\/p>\n
\u7b80\u5355\u603b\u7ed3\u4e3a\u4e00\u53e5\u8bdd：Agent\u5e94\u7528\u4e2d\u4e0a\u4e0b\u6587\u5de5\u7a0b\u5927\u6709\u53ef\u4e3a（\u4ecd\u6709\u5f88\u5927\u4f18\u5316\u7a7a\u95f4）\u3002<\/p>\n
1.3.2 Agent Loop<\/p>\n
\u57281.3.1\u5c0f\u8282\u4e2d\u6211\u4eec\u8ba8\u8bba\u4e86Agent\u6846\u67b6\u7684\u6838\u5fc3\u662f\u4e0a\u4e0b\u6587\u5de5\u7a0b，\u800c\u4e0a\u4e0b\u6587\u5de5\u7a0b\u7684\u6838\u5fc3\u5f15\u64ce\u6216\u8005\u79f0\u4e4b\u4e3a\u8fd0\u884c\u6846\u67b6\u5219\u662fAgent Loop\u3002<\/p>\n
Agent Loop\u4e5f\u4e0d\u795e\u79d8，\u672c\u8d28\u662f\u4e00\u4e2aWhile\u5faa\u73af，\u6bcf\u4e00\u6b21\u8fed\u4ee3\u662f\u4e00\u6b21LLM\u63a8\u7406\u5916\u52a0\u5de5\u5177\u8c03\u7528\u548c\u4e0a\u4e0b\u6587\u5904\u7406，\u4e5f\u5c31\u662f\u8bf4\u6240\u6709Agent\u884c\u4e3a\u7684\u53d1\u751f\u90fd\u662f\u5728\u8fd9\u4e2aWhile\u5faa\u73af\u91cc\u9762，\u76f4\u5230\u4efb\u52a1\u5b8c\u6210\u9000\u51fa\u3002<\/p>\n
![\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030531-69a25b7b5d866.png\")
<\/p>\n
\u5178\u578b\u7684\u5de5\u4f5c\u6d41\u7a0b\u5982\u4e0b：<\/p>\n
\u521d\u59cb\u4e0a\u4e0b\u6587（\u7cfb\u7edf\u63d0\u793a\u8bcd+\u7528\u6237\u8bf7\u6c42）
\n\u2193
\n[agent loop\u5f00\u59cb]
\n\u2193
\nagent\u8bfb\u53d6\u4e0a\u4e0b\u6587 \u2192 \u601d\u8003 \u2192 \u51b3\u5b9a\u884c\u52a8
\n\u2193
\n\u6267\u884c\u5de5\u5177\/\u884c\u52a8 \u2192 \u83b7\u5f97\u7ed3\u679c
\n\u2193
\n\u7ed3\u679c\u8ffd\u52a0\u5230\u4e0a\u4e0b\u6587
\n\u2193
\n[\u5faa\u73af\u7ee7\u7eed\u6216\u7ed3\u675f]<\/p>\n
\u7ec6\u5206\u5230While\u5faa\u73af\u5230\u6bcf\u4e00\u6b21\u8fed\u4ee3（Turn）\u53ef\u7b80\u5355\u8868\u793a\u4e3a：<\/p>\n
\u521d\u59cb\u5316\u4e0a\u4e0b\u6587（\u7528\u6237\u8bf7\u6c42）
\n\u2193
\n\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
\n\u2502 Agent Loop \u2502
\n\u2502 \u2502
\n\u2502 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u2502
\n\u2502 \u2502 Turn 1 \u2502 \u2502
\n\u2502 \u2502 LLM Call \u63a8\u7406 #1 \u2502
\n\u2502 \u2502 \u2192 \u89e3\u6790LLM\u54cd\u5e94 \u2502 \u2502
\n\u2502 \u2502 \u2192 \u6267\u884c\u5de5\u51771 \u2502 \u2502
\n\u2502 \u2502 \u2192 \u8fd4\u56de\u7ed3\u679c，\u66f4\u65b0\u4e0a\u4e0b\u6587 \u2502 \u2502
\n\u2502 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2502
\n\u2502 \u2193 \u2502
\n\u2502 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u2502
\n\u2502 \u2502 Turn 2 \u2502 \u2502
\n\u2502 \u2502 LLM Call \u63a8\u7406 #2 \u2502 \u2502
\n\u2502 \u2502 \u2192 \u6267\u884c\u5de5\u51772 \u2502 \u2502
\n\u2502 \u2502 \u2192 \u8fd4\u56de\u7ed3\u679c，\u66f4\u65b0\u4e0a\u4e0b\u6587 \u2502 \u2502
\n\u2502 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2502
\n\u2502 …. \u2502
\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518
\n\u2193
\n\u5b8c\u6210(\u5f53\u67d0\u4e00\u6b21Turn\u4e0d\u518d\u6267\u884c\u5de5\u5177\u5373\u8868\u793a\u5b8c\u6210，\u8fd4\u56de\u6240\u6709\u7ed3\u679c\u4fe1\u606f）<\/p>\n
Agent Loop\u901a\u8fc7\u5728\u6bcf\u6b21\u8fed\u4ee3\u4e2d\u8bfb\u53d6\u3001\u5229\u7528\u548c\u66f4\u65b0\u4e0a\u4e0b\u6587\u6765\u5b8c\u6210\u4efb\u52a1；\u4e0a\u4e0b\u6587\u5de5\u7a0b\u5219\u662f\u8bbe\u8ba1\u5982\u4f55\u7ec4\u7ec7\u3001\u7ba1\u7406\u548c\u4f18\u5316\u8fd9\u4e9b\u4e0a\u4e0b\u6587\u4fe1\u606f\u4ee5\u63d0\u5347Agent\u7684\u51b3\u7b56\u8d28\u91cf\u548c\u6548\u7387\u3002<\/p>\n
\u56de\u5230\u672c\u8282\u7684\u4e3b\u9898，\u53ef\u603b\u7ed3\u4e3a\u4e00\u53e5\u8bdd：Agent\u6846\u67b6\u8bbe\u8ba1\u7684\u6838\u5fc3\u5c31\u662f\u5728Agent Loop\u8fd9\u4e2aWhile\u5faa\u73af\u4e2d\u8bbe\u8ba1\u5982\u4f55\u7ba1\u7406\u4e0a\u4e0b\u6587\u3002<\/p>\n
\u627f\u63a5\u4e0a\u7bc7：Agent\u6846\u67b6\u8bbe\u8ba1\u7684\u6838\u5fc3\u662f\u5728Agent Loop\u8fd9\u4e2aWhile\u5faa\u73af\u4e2d\u8bbe\u8ba1\u5982\u4f55\u7ba1\u7406\u4e0a\u4e0b\u6587，\u672c\u7bc7\u5373\u56f4\u7ed5\u8fd9\u4e2a\u6838\u5fc3\u8bba\u70b9\u5c55\u5f00\u3002<\/p>\n
2.1 Agent \u6846\u67b6\u67b6\u6784\u56fe\u4e00\u89c8<\/p>\n
\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
\n\u2502 User Interface（CLI REPL Layer ） \u2502
\n\u2502 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u2502
\n\u2502 \u2502 User Input \u2502 \u2502 Exit\/ \u2502 \u2502 Message History \u2502 \u2502
\n\u2502 \u2502 Handler \u2502 \u2502 Clear Cmd \u2502 \u2502 Management \u2502 \u2502
\n\u2502 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2502
\n\u2502 \u2502 \u2502
\n\u2502 \u25bc \u2502
\n\u2502 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u2502
\n\u2502 \u2502 Agent Loop Core \u2502 \u2502
\n\u2502 \u2502 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u2502 \u2502
\n\u2502 \u2502 \u2502 LLM Call \u2502\u2500\u2500\u2500\u25b6\u2502 Tool Call \u2502\u2500\u2500\u2500\u25b6\u2502 Tool Exec \u2502 \u2502 \u2502
\n\u2502 \u2502 \u2502 (DeepSeek) \u2502 \u2502 Parser \u2502 \u2502 Engine \u2502 \u2502 \u2502
\n\u2502 \u2502 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2502 \u2502
\n\u2502 \u2502 \u2502 \u2502 \u2502 \u2502
\n\u2502 \u2502 \u2502\u25c0\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2502 \u2502
\n\u2502 \u2502 \u2502 (Tool Results Feedback) \u2502 \u2502
\n\u2502 \u2502 \u25bc \u2502 \u2502
\n\u2502 \u2502 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u2502 \u2502
\n\u2502 \u2502 \u2502 Response \u2502\u2500\u2500\u2500\u25b6\u2502 Context \u2502 \u2502 \u2502
\n\u2502 \u2502 \u2502 Formatter \u2502 \u2502 Manager \u2502 \u2502 \u2502
\n\u2502 \u2502 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2502 \u2502
\n\u2502 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2502
\n\u2502 \u2502 \u2502
\n\u2502 \u25bc \u2502
\n\u2502 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u2502
\n\u2502 \u2502 Tools Registry (TOOLS) \u2502 \u2502
\n\u2502 \u2502 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u2502 \u2502
\n\u2502 \u2502 \u2502 shell_ \u2502 \u2502 file_ \u2502 \u2502 file_ \u2502 \u2502 python_ \u2502 \u2502 \u2502
\n\u2502 \u2502 \u2502 exec \u2502 \u2502 read \u2502 \u2502 write \u2502 \u2502 exec \u2502 \u2502 \u2502
\n\u2502 \u2502 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2502 \u2502
\n\u2502 \u2502 \u2502 \u2502 \u2502 \u2502 \u2502 \u2502
\n\u2502 \u2502 \u25bc \u25bc \u25bc \u25bc \u2502 \u2502
\n\u2502 \u2502 [Function] [Function] [Function] [Function] \u2502 \u2502
\n\u2502 \u2502 [Schema] [Schema] [Schema] [Schema] \u2502 \u2502
\n\u2502 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2502
\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<\/p>\n
\u518dreview\u4e00\u4e0b\u5de5\u4f5c\u6d41：<\/p>\n
\u521d\u59cb\u4e0a\u4e0b\u6587（\u7cfb\u7edf\u63d0\u793a\u8bcd+\u7528\u6237\u8bf7\u6c42）
\n\u2193
\n[agent loop\u5f00\u59cb]
\n\u2193
\nagent\u8bfb\u53d6\u4e0a\u4e0b\u6587 \u2192 \u601d\u8003 \u2192 \u51b3\u5b9a\u884c\u52a8
\n\u2193
\n\u6267\u884c\u5de5\u5177\/\u884c\u52a8 \u2192 \u83b7\u5f97\u7ed3\u679c
\n\u2193
\n\u7ed3\u679c\u8ffd\u52a0\u5230\u4e0a\u4e0b\u6587
\n\u2193
\n[\u5faa\u73af\u7ee7\u7eed\u6216\u7ed3\u675f]<\/p>\n
2.2 Agent \u6846\u67b6\u4e09\u5927\u8981\u7d20\u8bbe\u8ba1<\/p>\n
2.2.1 LLM Call<\/p>\n
\u91c7\u7528\u6781\u7b80\u8bbe\u8ba1，\u4ee5DeepSeek\u6a21\u578b\u793a\u4f8b\u8bf4\u660e<\/p>\n
\u4e3a\u4fdd\u8bc1\u4ee3\u7801\u7684\u6700\u5927\u53ef\u8bfb\u6027，\u8fd9\u91cc\u4f7f\u7528\u540c\u6b65\u975e\u6d41\u5f0f\u8c03\u7528\u3002<\/p>\n
2.2.2 Tools Call<\/p>\n
\u91c7\u7528\u6781\u7b80\u7684\u5de5\u5177\u96c6，\u64cd\u4f5c\u5bf9\u8c61\u5305\u542b\u6587\u4ef6\u3001Shell\u548cPython\u4ee3\u7801\u6267\u884c<\/p>\n
1）Tools \u5b9e\u73b0：\u603b\u5171\u652f\u63014\u4e2a\u5de5\u5177\u51fd\u6570<\/p>\n
2）Tools \u6ce8\u518c：\u8fd9\u91cc\u9009\u62e9\u7684\u662f\u624b\u52a8\u7ef4\u62a4\u5b57\u5178\u6620\u5c04\u7684\u65b9\u5f0f name \u2192 (function, OpenAI function schema) ，\u8fd9\u4e00\u6b65\u662f\u4e3a\u4e86\u89e3\u6790llm call \u7684response\u65f6\u53ef\u4ee5\u6839\u636ename\u5339\u914d\u9700\u8981\u5177\u4f53\u6267\u884c\u54ea\u4e2atool<\/p>\n
Tools \u7684\u5b9a\u4e49\u9075\u5faa\u7684\u662f OpenAI Function Calling \u7684\u6807\u51c6\u683c\u5f0f（\u4e5f\u79f0 OpenAI Tools API schema）<\/p>\n
2.2.3 Context Engineering<\/p>\n
System Prompt：\u6781\u7b80\u7cfb\u7edf\u63d0\u793a\u8bcd，\u544a\u77e5LLM\u53ef\u7528\u5de5\u5177\u548cReAct\u601d\u8003\u65b9\u5f0f<\/p>\n<\/li>\n
\u7528\u6237Session\u7ba1\u7406：\u4f7f\u7528messages \u5217\u8868\u65b9\u5f0f（OpenAI chat \u683c\u5f0f），\u5b83\u662f\u6838\u5fc3\u72b6\u6001，\u7d2f\u79ef\u7cfb\u7edf\u63d0\u793a\u8bcd\u3001\u7528\u6237\u6d88\u606f\u3001\u52a9\u624b\u54cd\u5e94\u548c\u5de5\u5177\u7ed3\u679c<\/p>\n
2.3 Agent \u6846\u67b6\u4ee3\u7801\u5b9e\u73b0<\/p>\n
2.3.1 \u7b2c\u4e00\u90e8\u5206：Agent Loop \u4e0e \u4e0a\u4e0b\u6587<\/p>\n<\/li>\n
\u57fa\u7840\u6d41\u7a0b： LLM call \u2192 parse tool_calls \u2192 execute \u2192 append results to messages \u2192 loop or exit<\/p>\n<\/li>\n
\u5b89\u5168\u8bbe\u7f6e：\u4e3awhile\u5faa\u73af\u8bbe\u7f6e\u4e86\u4e00\u4e2a\u8fed\u4ee3\u7684\u5b89\u5168\u4e0a\u9650：20 \u8f6e（MAX_TURNS=200）<\/p>\n<\/li>\n
\u4f7f\u7528\u5168\u5c40\u53d8\u91cfmessage\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u7684\u8f7d\u4f53，\u7d2f\u79ef\u7cfb\u7edf\u63d0\u793a\u8bcd\u3001\u7528\u6237\u6d88\u606f\u3001\u52a9\u624b\u54cd\u5e94\u548c\u5de5\u5177\u7ed3\u679c<\/p>\n<\/li>\n<\/ul>\n
\u5176\u4e2d，\u53d8\u91cfmessage\u6309\u5982\u4e0b\u89c4\u5219\u66f4\u65b0<\/p>\n
# ============================================================
\n# Agent Loop \u2014 \u6838\u5fc3
\n# ============================================================
\nMAX_TURNS = 20
\ndef agent_loop(user_message: str, messages: list, client: OpenAI) -> str:
\n"""
\nAgent Loop：while \u5faa\u73af\u9a71\u52a8 LLM \u63a8\u7406\u4e0e\u5de5\u5177\u8c03\u7528\u3002
\n\u6d41\u7a0b：
\n1. \u5c06\u7528\u6237\u6d88\u606f\u8ffd\u52a0\u5230 messages
\n2. \u8c03\u7528 LLM
\n3. \u82e5 LLM \u8fd4\u56de tool_calls \u2192 \u9010\u4e2a\u6267\u884c \u2192 \u7ed3\u679c\u8ffd\u52a0\u5230 messages \u2192 \u7ee7\u7eed\u5faa\u73af
\n4. \u82e5 LLM \u76f4\u63a5\u8fd4\u56de\u6587\u672c（\u65e0 tool_calls）\u2192 \u9000\u51fa\u5faa\u73af，\u8fd4\u56de\u6587\u672c
\n5. \u5b89\u5168\u4e0a\u9650 MAX_TURNS \u8f6e
\n"""
\nmessages.append({"role": "user", "content": user_message})
\ntool_schemas = [t["schema"] for t in TOOLS.values()]
\nfor turn in range(1, MAX_TURNS + 1):
\n# — LLM Call —
\nresponse = client.chat.completions.create(
\nmodel="deepseek-chat",
\nmessages=messages,
\ntools=tool_schemas,
\n)
\nchoice = response.choices[0]
\nassistant_msg = choice.message
\n# \u5c06 assistant \u6d88\u606f\u8ffd\u52a0\u5230\u4e0a\u4e0b\u6587
\nmessages.append(assistant_msg.model_dump())
\n# — \u7ec8\u6b62\u6761\u4ef6：\u65e0 tool_calls —
\nif not assistant_msg.tool_calls:
\nreturn assistant_msg.content or ""
\n# — \u6267\u884c\u6bcf\u4e2a tool_call —
\nfor tool_call in assistant_msg.tool_calls:
\nname = tool_call.function.name
\nraw_args = tool_call.function.arguments
\nprint(f" [tool] {name}({raw_args})")
\n# \u89e3\u6790\u53c2\u6570\u5e76\u8c03\u7528\u5de5\u5177
\ntry:
\nargs = json.loads(raw_args)
\nexcept json.JSONDecodeError:
\nargs = {}
\ntool_entry = TOOLS.get(name)
\nif tool_entry is None:
\nresult = f"[error] unknown tool: {name}"
\nelse:
\nresult = tool_entry["function"](**args)
\n# \u5c06\u5de5\u5177\u7ed3\u679c\u8ffd\u52a0\u5230\u4e0a\u4e0b\u6587
\nmessages.append(
\n{
\n"role": "tool",
\n"tool_call_id": tool_call.id,
\n"content": result,
\n}
\n)
\nreturn "[agent] reached maximum turns, stopping."<\/p>\n
\u6ce8：\u8fd9\u91cc\u4f7f\u7528\u7684\u6a21\u578b\u4e3adeepseek-chat，\u4e3b\u8981\u8003\u91cf\u56e0\u7d20\u662f\u6a21\u578b\u652f\u6301Tool Calls，\u5e76\u4e14\u5b8c\u5168\u517c\u5bb9OpenAI\u7684SDK\u3002<\/p>\n
2.3.2 \u7b2c\u4e8c\u90e8\u5206：Tools \u5b9e\u73b0\u4e0e\u6ce8\u518c<\/p>\n
\u8fd9\u91cc\u4e3b\u8981\u5b9e\u73b0\u56db\u4e2a\u5de5\u5177\u51fd\u6570: shell_exec, file_read, file_write, python_exec<\/p>\n
# ============================================================
\n# Tools \u5b9e\u73b0 \u2014 4 \u4e2a\u5de5\u5177\u51fd\u6570
\n# ============================================================
\ndef shell_exec(command: str) -> str:
\n"""\u6267\u884c shell \u547d\u4ee4\u5e76\u8fd4\u56de stdout + stderr\u3002"""
\ntry:
\nresult = subprocess.run(
\ncommand,
\nshell=True,
\ncapture_output=True,
\ntext=True,
\ntimeout=30,
\n)
\noutput = result.stdout
\nif result.stderr:
\noutput += "\\\\n[stderr]\\\\n" + result.stderr
\nif result.returncode != 0:
\noutput += f"\\\\n[exit code: {result.returncode}]"
\nreturn output.strip() or "(no output)"
\nexcept subprocess.TimeoutExpired:
\nreturn "[error] command timed out after 30s"
\nexcept Exception as e:
\nreturn f"[error] {e}"
\ndef file_read(path: str) -> str:
\n"""\u8bfb\u53d6\u6587\u4ef6\u5185\u5bb9\u3002"""
\ntry:
\nwith open(path, "r", encoding="utf-8") as f:
\nreturn f.read()
\nexcept Exception as e:
\nreturn f"[error] {e}"
\ndef file_write(path: str, content: str) -> str:
\n"""\u5c06\u5185\u5bb9\u5199\u5165\u6587\u4ef6（\u81ea\u52a8\u521b\u5efa\u7236\u76ee\u5f55）\u3002"""
\ntry:
\nos.makedirs(os.path.dirname(path) or ".", exist_ok=True)
\nwith open(path, "w", encoding="utf-8") as f:
\nf.write(content)
\nreturn f"OK \u2014 wrote {len(content)} chars to {path}"
\nexcept Exception as e:
\nreturn f"[error] {e}"
\ndef python_exec(code: str) -> str:
\n"""\u5728\u5b50\u8fdb\u7a0b\u4e2d\u6267\u884c Python \u4ee3\u7801\u5e76\u8fd4\u56de\u8f93\u51fa\u3002"""
\ntry:
\nwith tempfile.NamedTemporaryFile(
\nmode="w", suffix=".py", delete=False, encoding="utf-8"
\n) as tmp:
\ntmp.write(code)
\ntmp_path = tmp.name
\nresult = subprocess.run(
\n[sys.executable, tmp_path],
\ncapture_output=True,
\ntext=True,
\ntimeout=30,
\n)
\noutput = result.stdout
\nif result.stderr:
\noutput += "\\\\n[stderr]\\\\n" + result.stderr
\nreturn output.strip() or "(no output)"
\nexcept subprocess.TimeoutExpired:
\nreturn "[error] execution timed out after 30s"
\nexcept Exception as e:
\nreturn f"[error] {e}"
\nfinally:
\ntry:
\nos.unlink(tmp_path)
\nexcept OSError:
\npass<\/p>\n
\u5de5\u5177\u5b9e\u73b0\u597d\u4e86\u4e0b\u4e00\u6b65\u5c31\u662f\u6ce8\u518c，\u65b9\u4fbfAgent Loop\u53ef\u4ee5\u6839\u636eLLM\u7684\u8fd4\u56de\u7ed3\u679c\u6267\u884c\u5177\u4f53\u7684\u5de5\u5177\u65b9\u6cd5（\u5b9e\u9645\u4e0a\u5c31\u662f\u4e00\u4e2a\u5b57\u5178\u6620\u5c04name \u2192 {function, OpenAI schema}）<\/p>\n
# ============================================================
\n# Tools \u6ce8\u518c \u2014 name \u2192 (function, OpenAI function schema)
\n# ============================================================
\nTOOLS = {
\n"shell_exec": {
\n"function": shell_exec,
\n"schema": {
\n"type": "function",
\n"function": {
\n"name": "shell_exec",
\n"description": "Execute a shell command and return its output.",
\n"parameters": {
\n"type": "object",
\n"properties": {
\n"command": {
\n"type": "string",
\n"description": "The shell command to execute.",
\n}
\n},
\n"required": ["command"],
\n},
\n},
\n},
\n},
\n"file_read": {
\n"function": file_read,
\n"schema": {
\n"type": "function",
\n"function": {
\n"name": "file_read",
\n"description": "Read the contents of a file at the given path.",
\n"parameters": {
\n"type": "object",
\n"properties": {
\n"path": {
\n"type": "string",
\n"description": "Absolute or relative file path.",
\n}
\n},
\n"required": ["path"],
\n},
\n},
\n},
\n},
\n"file_write": {
\n"function": file_write,
\n"schema": {
\n"type": "function",
\n"function": {
\n"name": "file_write",
\n"description": "Write content to a file (creates parent directories if needed).",
\n"parameters": {
\n"type": "object",
\n"properties": {
\n"path": {
\n"type": "string",
\n"description": "Absolute or relative file path.",
\n},
\n"content": {
\n"type": "string",
\n"description": "Content to write.",
\n},
\n},
\n"required": ["path", "content"],
\n},
\n},
\n},
\n},
\n"python_exec": {
\n"function": python_exec,
\n"schema": {
\n"type": "function",
\n"function": {
\n"name": "python_exec",
\n"description": "Execute Python code in a subprocess and return its output.",
\n"parameters": {
\n"type": "object",
\n"properties": {
\n"code": {
\n"type": "string",
\n"description": "Python source code to execute.",
\n}
\n},
\n"required": ["code"],
\n},
\n},
\n},
\n},
\n}<\/p>\n
Tools \u7684\u5b9a\u4e49\u9075\u5faa\u7684\u662f OpenAI Function Calling \u7684\u6807\u51c6\u683c\u5f0f（\u4e5f\u79f0 OpenAI Tools API schema）\u3002<\/p>\n
\u5177\u4f53\u6765\u8bf4，Agent \u4e2d\u6bcf\u4e2a\u5de5\u5177\u7684 schema \u5b57\u6bb5\u7684\u7ed3\u6784\u5982\u4e0b：<\/p>\n
{
\n"type": "function",
\n"function": {
\n"name": "…",
\n"description": "…",
\n"parameters": {
\n"type": "object",
\n"properties": { … },
\n"required": [ … ],
\n},
\n},
\n}<\/p>\n
2.3.3 \u7b2c\u4e09\u90e8\u5206：System Prompt<\/p>\n
\u8fd9\u91cc\u5b9a\u4e49System Prompt，\u6bcf\u4e00\u6b21\u4e0eLLM\u4ea4\u4e92\u90fd\u9700\u8981\u5e26\u4e0a\u5b83\u3002<\/p>\n
# ============================================================
\n# System Prompt
\n# ============================================================
\nSYSTEM_PROMPT = """You are a helpful AI assistant with access to the following tools:
\n1. shell_exec \u2014 run shell commands
\n2. file_read \u2014 read file contents
\n3. file_write \u2014 write content to a file
\n4. python_exec \u2014 execute Python code
\nThink step by step. Use tools when you need to interact with the file system, \\\\
\nrun commands, or execute code. When the task is complete, respond directly \\\\
\nwithout calling any tool."""<\/p>\n
\u660e\u786e\u544a\u77e5：\u4f60\u662f\u4e00\u4e2aAI\u52a9\u624b，\u5f53\u9700\u8981\u7684\u65f6\u5019\u53ef\u4ee5\u4f7f\u7528\u54ea\u4e9b\u5de5\u5177\u3002<\/p>\n
\u81f3\u6b64\u4e00\u4e2a\u6781\u7b80\u7684Agent\u6846\u67b6\u5c31\u6b64\u5b9e\u73b0\u5b8c\u6210，\u5355\u6587\u4ef6\u641e\u5b9a，\u5168\u90e8\u4ee3\u7801279\u884c\u3002<\/p>\n
2.4 \u57fa\u4e8e\u6781\u7b80 Agent \u6846\u67b6\u7684\u6781\u7b80 Agent \u5e94\u7528<\/p>\n
2.4.1 \u7528\u6237\u4ea4\u4e92\u754c\u9762\u8bbe\u8ba1 – Python CLI REPL<\/p>\n
\u6846\u67b6\u5b9e\u73b0\u5b8c\u6210\u4e4b\u540e，\u8ddd\u79bbAgent\u5e94\u7528\u5c31\u5269\u4e0b\u6700\u540e\u4e00\u4e2a\u7528\u6237\u4ea4\u4e92\u754c\u9762\u4e86\u3002<\/p>\n
\u907f\u514d\u589e\u52a0\u8bfb\u8005\u8ba4\u77e5\u8d1f\u62c5，\u4ece\u6781\u7b80\u601d\u60f3\u51fa\u53d1，\u8fd9\u91cc\u4f7f\u7528Python CLI REPL \u5373Python\u7684\u4ea4\u4e92\u5f0f\u547d\u4ee4\u884c\u4f5c\u4e3aAgent\u7684\u5165\u53e3：<\/p>\n
def main():
\napi_key = os.environ.get("DEEPSEEK_API_KEY")
\nif not api_key:
\nprint("Error: please set DEEPSEEK_API_KEY environment variable.")
\nsys.exit(1)
\nclient = OpenAI(api_key=api_key, base_url="https:\/\/api.deepseek.com")
\nmessages: list = [{"role": "system", "content": SYSTEM_PROMPT}]
\nprint("Agent ready. Type your message (or 'exit' to quit, 'clear' to reset).\\\\n")
\nwhile True:
\ntry:
\nuser_input = input("You> ").strip()
\nexcept (EOFError, KeyboardInterrupt):
\nprint("\\\\nBye.")
\nbreak
\nif not user_input:
\ncontinue
\nif user_input.lower() == "exit":
\nprint("Bye.")
\nbreak
\nif user_input.lower() == "clear":
\nmessages.clear()
\nmessages.append({"role": "system", "content": SYSTEM_PROMPT})
\nprint("(context cleared)\\\\n")
\ncontinue
\nreply = agent_loop(user_input, messages, client)
\nprint(f"\\\\nAgent> {reply}\\\\n")<\/p>\n
2.4.2 DeepSeek \u6ce8\u518c，\u83b7\u53d6 API Key<\/p>\n
\u7531\u4e8e\u672c\u6587Agent\u6846\u67b6\u7684LLM Provider\u662f\u57fa\u4e8eDeepSeek\u5b9e\u73b0\u7684，\u6240\u4ee5\u9700\u8981\u83b7\u53d6DeepSeek\u6a21\u578b（deepseek-chat\u6a21\u578b）\u7684API key\u624d\u80fd\u4f7f\u7528\u3002<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030532-69a25b7c99515.png\")
<\/p>\n
2.4.3 \u6781\u7b80 Agent \u5e94\u7528\u4f53\u9a8c<\/p>\n
\u4f7f\u7528\u4e4b\u524d\u8bbe\u7f6eAPI key<\/p>\n
export DEEPSEEK_API_KEY="sk-xxxxx"<\/p>\n
\u5148\u6765\u95ee\u5019\u4e00\u4e0b：<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030532-69a25b7cce4fb.png\")
<\/p>\n
\u8f93\u51fa\u7684\u57fa\u672c\u4e0a\u662fSystem Prompt\u4e2d\u7684\u5185\u5bb9，\u7b26\u5408\u9884\u671f\u3002<\/p>\n
1）\u7b2c1\u4e2a\u95ee\u9898<\/p>\n
\u5e2e\u6211\u67e5\u4e00\u4e0b\u5f53\u524d\u76ee\u5f55\u90fd\u6709\u54ea\u4e9b\u6587\u4ef6<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030533-69a25b7d06b45.png\")
<\/p>\n
2）\u7b2c2\u4e2a\u95ee\u9898<\/p>\n
\u8ba9\u5b83\u6267\u884c\u4e00\u4e2a\u7edf\u8ba1\u4efb\u52a1：\u5e2e\u6211\u7edf\u8ba1\u4e0b\u5f53\u524d\u76ee\u5f55\u4e0b\u7684\u4ee3\u7801\u884c\u6570\u4ee5\u53catoken\u6570<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030533-69a25b7d85027.png\")
<\/p>\n
\u53ef\u4ee5\u770b\u5230Agent Loop\u91cc\u5728\u6301\u7eed\u5730\u8c03\u7528Tools\u3001\u5199\u4ee3\u7801\u4ee5\u53ca\u6267\u884c\u4ee3\u7801<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030533-69a25b7dca758.png\")
<\/p>\n
\u6700\u7ec8\u7684\u7edf\u8ba1\u7ed3\u679c：<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030534-69a25b7e22076.png\")
<\/p>\n
\u505a\u5b8c\u4efb\u52a1，\u4f7f\u7528exit\u547d\u4ee4\u9000\u51fa\u5f53\u524dsession<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030534-69a25b7e89821.png\")
<\/p>\n
\u53ef\u4ee5\u770b\u5230\u5b9e\u73b0\u7684Agent\u5e94\u7528，\u867d\u7136\u5b9e\u73b0\u6781\u7b80，\u4f46\u662f\u529f\u80fd\u53ef\u4ee5\u4e00\u70b9\u4e0d\u7b80\u5355（\u5f53Agent\u62e5\u6709\u6587\u4ef6\u8bfb\u5199\u6743\u9650，\u5916\u52a0Shell\u5de5\u5177\u4ee5\u53ca\u4ee3\u7801\u751f\u6210\u4e0e\u6267\u884c\u6743\u9650，\u5b83\u5728\u672c\u673a\u4e0a\u771f\u7684\u53ef\u4ee5"\u4e3a\u6240\u6b32\u4e3a")\u3002<\/p>\n
\u8981\u77e5\u9053OpenClaw\u7684\u5e95\u5c42Agent Core（Pi Agent）\u7684Tools\u5c42\u4e5f\u662f\u6709\u4e14\u4ec5\u5305\u542b\u56db\u4e2a\u5de5\u5177\u65b9\u6cd5：\u8bfb\u6587\u4ef6（Read）\u3001\u5199\u6587\u4ef6（Write）\u3001\u7f16\u8f91\u6587\u4ef6（Edit）\u3001\u547d\u4ee4\u884c（Shell），\u5176\u4ed6\u7684\u4e30\u5bcc\u4e14\u5f3a\u5927\u80fd\u529b\u5747\u9760\u4e8b\u4ef6\u673a\u5236\u53caSkills\u6269\u5c55\u800c\u6765\u3002<\/p>\n
\u9886\u53d6\u65b9\u5f0f\u5728\u6587\u672b<\/p>\n
\u76ee\u524dAI\u5927\u6a21\u578b\u7684\u6280\u672f\u5c97\u4f4d\u4e0e\u80fd\u529b\u57f9\u517b\u968f\u7740\u4eba\u5de5\u667a\u80fd\u6280\u672f\u7684\u8fc5\u901f\u53d1\u5c55\u548c\u5e94\u7528 ， \u5927\u6a21\u578b\u4f5c\u4e3a\u5176\u4e2d\u7684\u91cd\u8981\u7ec4\u6210\u90e8\u5206 ， \u6b63\u9010\u6e10\u6210\u4e3a\u63a8\u52a8\u4eba\u5de5\u667a\u80fd\u53d1\u5c55\u7684\u91cd\u8981\u5f15\u64ce \u3002\u5927\u6a21\u578b\u4ee5\u5176\u5f3a\u5927\u7684\u6570\u636e\u5904\u7406\u548c\u6a21\u5f0f\u8bc6\u522b\u80fd\u529b， \u5e7f\u6cdb\u5e94\u7528\u4e8e\u81ea\u7136\u8bed\u8a00\u5904\u7406 \u3001\u8ba1\u7b97\u673a\u89c6\u89c9 \u3001 \u667a\u80fd\u63a8\u8350\u7b49\u9886\u57df ，\u4e3a\u5404\u884c\u5404\u4e1a\u5e26\u6765\u4e86\u9769\u547d\u6027\u7684\u6539\u53d8\u548c\u673a\u9047 \u3002<\/p>\n
\u76ee\u524d，\u5f00\u6e90\u4eba\u5de5\u667a\u80fd\u5927\u6a21\u578b\u5df2\u5e94\u7528\u4e8e\u533b\u7597\u3001\u653f\u52a1\u3001\u6cd5\u5f8b\u3001\u6c7d\u8f66\u3001\u5a31\u4e50\u3001\u91d1\u878d\u3001\u4e92\u8054\u7f51\u3001\u6559\u80b2\u3001\u5236\u9020\u4e1a\u3001\u4f01\u4e1a\u670d\u52a1\u7b49\u591a\u4e2a\u573a\u666f，\u5176\u4e2d，\u5e94\u7528\u4e8e\u91d1\u878d\u3001\u4f01\u4e1a\u670d\u52a1\u3001\u5236\u9020\u4e1a\u548c\u6cd5\u5f8b\u9886\u57df\u7684\u5927\u6a21\u578b\u5728\u672c\u6b21\u8c03\u7814\u4e2d\u5360\u6bd4\u8d85\u8fc7 30%\u3002
\n![\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030534-69a25b7e9ed59.png\")
<\/p>\n
\u968f\u7740AI\u5927\u6a21\u578b\u6280\u672f\u7684\u8fc5\u901f\u53d1\u5c55，\u76f8\u5173\u5c97\u4f4d\u7684\u9700\u6c42\u4e5f\u65e5\u76ca\u589e\u52a0\u3002\u5927\u6a21\u578b\u4ea7\u4e1a\u94fe\u50ac\u751f\u4e86\u4e00\u6279\u9ad8\u85aa\u65b0\u804c\u4e1a：
\n![\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030535-69a25b7f0c16b.png\")
<\/p>\n
\u4eba\u5de5\u667a\u80fd\u5927\u6f6e\u5df2\u6765，\u4e0d\u52a0\u5165\u5c31\u53ef\u80fd\u88ab\u6dd8\u6c70\u3002\u5982\u679c\u4f60\u662f\u6280\u672f\u4eba，\u5c24\u5176\u662f\u4e92\u8054\u7f51\u4ece\u4e1a\u8005，\u73b0\u5728\u5c31\u5f00\u59cb\u5b66\u4e60AI\u5927\u6a21\u578b\u6280\u672f，\u771f\u7684\u662f\u7ed9\u4f60\u7684\u4eba\u751f\u4e00\u4e2a\u91cd\u8981\u5efa\u8bae！<\/p>\n
\u53ea\u8981\u4f60\u771f\u5fc3\u60f3\u5b66\u4e60AI\u5927\u6a21\u578b\u6280\u672f，\u8fd9\u4efd\u7cbe\u5fc3\u6574\u7406\u7684\u5b66\u4e60\u8d44\u6599\u6211\u613f\u610f\u65e0\u507f\u5206\u4eab\u7ed9\u4f60，\u4f46\u662f\u60f3\u5b66\u6280\u672f\u53bb\u4e71\u641e\u7684\u4eba\u522b\u6765\u627e\u6211！<\/p>\n
\u5728\u5f53\u524d\u8fd9\u4e2a\u4eba\u5de5\u667a\u80fd\u9ad8\u901f\u53d1\u5c55\u7684\u65f6\u4ee3，AI\u5927\u6a21\u578b\u6b63\u5728\u6df1\u523b\u6539\u53d8\u5404\u884c\u5404\u4e1a\u3002\u6211\u56fd\u5bf9\u9ad8\u6c34\u5e73AI\u4eba\u624d\u7684\u9700\u6c42\u4e5f\u65e5\u76ca\u589e\u957f，\u771f\u6b63\u61c2\u6280\u672f\u3001\u80fd\u843d\u5730\u7684\u4eba\u624d\u4f9d\u65e7\u7d27\u7f3a\u3002\u6211\u4e5f\u5e0c\u671b\u901a\u8fc7\u8fd9\u4efd\u8d44\u6599，\u80fd\u591f\u5e2e\u52a9\u66f4\u591a\u6709\u5fd7\u4e8eAI\u9886\u57df\u7684\u670b\u53cb\u5165\u95e8\u5e76\u6df1\u5165\u5b66\u4e60\u3002<\/p>\n
\u771f\u8bda\u65e0\u507f\u5206\u4eab！！！
\nvx\u626b\u63cf\u4e0b\u65b9\u4e8c\u7ef4\u7801\u5373\u53ef
\n\u52a0\u4e0a\u540e\u4f1a\u4e00\u4e2a\u4e2a\u7ed9\u5927\u5bb6\u53d1
\n\u3010\u9644\u8d60\u4e00\u8282\u514d\u8d39\u7684\u76f4\u64ad\u8bb2\u5ea7，\u6280\u672f\u5927\u4f6c\u5e26\u4f60\u5b66\u4e60\u5927\u6a21\u578b\u7684\u76f8\u5173\u77e5\u8bc6\u3001\u5b66\u4e60\u601d\u8def\u3001\u5c31\u4e1a\u524d\u666f\u4ee5\u53ca\u600e\u4e48\u7ed3\u5408\u5f53\u524d\u7684\u5de5\u4f5c\u53d1\u5c55\u65b9\u5411\u7b49，\u6b22\u8fce\u5927\u5bb6~\u3011
\n![\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030535-69a25b7f6eb50.png\")
<\/p>\n
\u81ea\u6211\u4eec\u4e0eMoPaaS\u9b54\u6cca\u4e91\u5408\u4f5c\u4ee5\u6765，\u6211\u4eec\u4e0d\u65ad\u6253\u78e8\u8bfe\u7a0b\u4f53\u7cfb\u4e0e\u6280\u672f\u5185\u5bb9，\u5728\u7ec6\u8282\u4e0a\u7cbe\u76ca\u6c42\u7cbe，\u540c\u65f6\u5728\u6280\u672f\u5c42\u9762\u4e5f\u65b0\u589e\u4e86\u8bb8\u591a\u524d\u6cbf\u4e14\u5b9e\u7528\u7684\u5185\u5bb9，\u529b\u6c42\u4e3a\u5927\u5bb6\u5e26\u6765\u66f4\u7cfb\u7edf\u3001\u66f4\u5b9e\u6218\u3001\u66f4\u843d\u5730\u7684\u5927\u6a21\u578b\u5b66\u4e60\u4f53\u9a8c\u3002<\/p>\n
![\"\u56fe\u7247\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030537-69a25b814c02c.gif\")
<\/p>\n
\u5e0c\u671b\u8fd9\u4efd\u7cfb\u7edf\u3001\u5b9e\u7528\u7684\u5927\u6a21\u578b\u5b66\u4e60\u8def\u5f84，\u80fd\u591f\u5e2e\u52a9\u4f60\u4ece\u96f6\u5165\u95e8，\u8fdb\u9636\u5230\u5b9e\u6218，\u771f\u6b63\u638c\u63e1AI\u65f6\u4ee3\u7684\u6838\u5fc3\u6280\u80fd！<\/p>\n
![\"\u56fe\u7247\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030537-69a25b816b8d1.png\")
<\/p>\n
\u4ece\u96f6\u5230\u7cbe\u901a\u5b8c\u6574\u95ed\u73af：\u3010\u57fa\u7840\u7406\u8bba \u2192RAG\u5f00\u53d1 \u2192 Agent\u8bbe\u8ba1 \u2192 \u6a21\u578b\u5fae\u8c03\u4e0e\u79c1\u6709\u5316\u90e8\u7f72\u8c03\u2192\u70ed\u95e8\u6280\u672f\u30115\u5927\u6a21\u5757，\u5185\u5bb9\u6bd4\u4f20\u7edf\u6559\u6750\u66f4\u8d34\u8fd1\u4f01\u4e1a\u5b9e\u6218！<\/p>\n<\/li>\n
\u5927\u91cf\u771f\u5b9e\u9879\u76ee\u6848\u4f8b： \u5e26\u4f60\u4eb2\u81ea\u4e0a\u624b\u641e\u6570\u636e\u6e05\u6d17\u3001\u6a21\u578b\u8c03\u4f18\u8fd9\u4e9b\u786c\u6838\u64cd\u4f5c，\u628a\u8bfe\u672c\u77e5\u8bc6\u53d8\u6210\u771f\u672c\u4e8b\u200c！<\/p>\n<\/li>\n<\/ul>\n
\u5e94\u5c4a\u6bd5\u4e1a\u751f\u200c： \u65e0\u5de5\u4f5c\u7ecf\u9a8c\u4f46\u60f3\u8981\u7cfb\u7edf\u5b66\u4e60AI\u5927\u6a21\u578b\u6280\u672f，\u671f\u5f85\u901a\u8fc7\u5b9e\u6218\u9879\u76ee\u638c\u63e1\u6838\u5fc3\u6280\u672f\u3002<\/p>\n
\u96f6\u57fa\u7840\u8f6c\u578b\u200c： \u975e\u6280\u672f\u80cc\u666f\u4f46\u5173\u6ce8AI\u5e94\u7528\u573a\u666f，\u8ba1\u5212\u901a\u8fc7\u4f4e\u4ee3\u7801\u5de5\u5177\u5b9e\u73b0\u201cAI+\u884c\u4e1a\u201d\u8de8\u754c\u200c\u3002<\/p>\n
\u4e1a\u52a1\u8d4b\u80fd\u7a81\u7834\u74f6\u9888： \u4f20\u7edf\u5f00\u53d1\u8005（Java\/\u524d\u7aef\u7b49）\u5b66\u4e60Transformer\u67b6\u6784\u4e0eLangChain\u6846\u67b6，\u5411AI\u5168\u6808\u5de5\u7a0b\u5e08\u8f6c\u578b\u200c\u3002<\/p>\n
![\"image.png\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030537-69a25b81ddf09.png\")
<\/p>\n
vx\u626b\u63cf\u4e0b\u65b9\u4e8c\u7ef4\u7801\u5373\u53ef
\n\u3010\u9644\u8d60\u4e00\u8282\u514d\u8d39\u7684\u76f4\u64ad\u8bb2\u5ea7，\u6280\u672f\u5927\u4f6c\u5e26\u4f60\u5b66\u4e60\u5927\u6a21\u578b\u7684\u76f8\u5173\u77e5\u8bc6\u3001\u5b66\u4e60\u601d\u8def\u3001\u5c31\u4e1a\u524d\u666f\u4ee5\u53ca\u600e\u4e48\u7ed3\u5408\u5f53\u524d\u7684\u5de5\u4f5c\u53d1\u5c55\u65b9\u5411\u7b49，\u6b22\u8fce\u5927\u5bb6~\u3011
\n<\/p>\n
\u672c\u6559\u7a0b\u6bd4\u8f83\u73cd\u8d35，\u4ec5\u9650\u5927\u5bb6\u81ea\u884c\u5b66\u4e60，\u4e0d\u8981\u4f20\u64ad！\u66f4\u4e25\u7981\u5546\u7528！<\/p>\n
\u5927\u6a21\u578b\u5b66\u4e60\u8def\u7ebf\u56fe，\u6574\u4f53\u5206\u4e3a5\u4e2a\u5927\u7684\u9636\u6bb5：
\n![\"\u56fe\u7247\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030538-69a25b822a1ca.png\")
<\/p>\n
![\"\u56fe\u7247\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030540-69a25b84a78aa.png\")
<\/p>\n
\u4ece0\u5230\u638c\u63e1\u4e3b\u6d41\u5927\u6a21\u578b\u6280\u672f\u89c6\u9891\u6559\u7a0b（\u6db5\u76d6\u6a21\u578b\u8bad\u7ec3\u3001\u5fae\u8c03\u3001RAG\u3001LangChain\u3001Agent\u5f00\u53d1\u7b49\u5b9e\u6218\u65b9\u5411）<\/p>\n
![\"\u56fe\u7247\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260228030540-69a25b84c2119.png\")
<\/p>\n
\u65b0\u624b\u5fc5\u5907\u7684\u5927\u6a21\u578b\u5b66\u4e60PDF\u4e66\u5355\u6765\u4e86！\u5168\u662f\u786c\u6838\u77e5\u8bc6，\u5e2e\u4f60\u5c11\u8d70\u5f2f\u8def（\u4e0d\u5439\u725b，\u771f\u6709\u7528）
\n![\"\u56fe\u7247\"](\"2026-02-284mujitguu3a.png\")
<\/p>\n
\u6536\u96c670+\u62a5\u544a\u4e0e\u767d\u76ae\u4e66，\u4e86\u89e3\u884c\u4e1a\u6700\u65b0\u52a8\u6001！
\n![\"\u56fe\u7247\"](\"2026-02-28iulokvsb0br.png\")
<\/p>\n
AI\u5927\u6a21\u578b\u5c97\u4f4d\u9762\u8bd5\u7ecf\u9a8c\u603b\u7ed3（\u8c01\u5b66\u6280\u672f\u4e0d\u662f\u4e3a\u4e86\u8d5a$\u5462，\u627e\u4e2a\u597d\u7684\u5c97\u4f4d\u5f88\u91cd\u8981）![\"\u56fe\u7247\"](\"2026-02-281ydn3qhiip2.gif\")
\n![\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"](\"2026-02-285lbb22f0h1s.png\")
<\/p>\n
![\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"](\"2026-02-2813jcdypbrpv.png\")
<\/p>\n
\u7531\u4e8e\u7bc7\u5e45\u6709\u9650<\/p>\n
\u53ea\u5c55\u793a\u90e8\u5206\u8d44\u6599<\/p>\n
\u5e76\u4e14\u8fd8\u5728\u6301\u7eed\u66f4\u65b0\u4e2d\u2026<\/p>\n
\u771f\u8bda\u65e0\u507f\u5206\u4eab！！！
\nvx\u626b\u63cf\u4e0b\u65b9\u4e8c\u7ef4\u7801\u5373\u53ef
\n\u52a0\u4e0a\u540e\u4f1a\u4e00\u4e2a\u4e2a\u7ed9\u5927\u5bb6\u53d1
\n\u3010\u9644\u8d60\u4e00\u8282\u514d\u8d39\u7684\u76f4\u64ad\u8bb2\u5ea7，\u6280\u672f\u5927\u4f6c\u5e26\u4f60\u5b66\u4e60\u5927\u6a21\u578b\u7684\u76f8\u5173\u77e5\u8bc6\u3001\u5b66\u4e60\u601d\u8def\u3001\u5c31\u4e1a\u524d\u666f\u4ee5\u53ca\u600e\u4e48\u7ed3\u5408\u5f53\u524d\u7684\u5de5\u4f5c\u53d1\u5c55\u65b9\u5411\u7b49，\u6b22\u8fce\u5927\u5bb6~\u3011
\n<\/p>\n","protected":false},"excerpt":{"rendered":"
\u5e74\u521d\u706b\u7206\u73b0\u5728\u70ed\u5ea6\u4f9d\u65e7\u7684OpenClaw\u4e3aAI Agent\u5e26\u6765\u4e86\u65b0\u7684\u60f3\u8c61\u3002\u5982\u679c\u8bf42025\u662fAI Agent\u667a\u80fd\u4f53\u5143\u5e74，\u60f3\u5fc52026\u5c06\u4f1a\u662fAI Agent\u771f\u6b63\u5546\u7528\u5316\u7684\u5f00\u7aef，\u800cAI Agent\u5546\u7528\u5316\u7684\u524d\u63d0\u662f\u5404\u884c\u5404\u4e1a\u5f00\u59cb\u843d\u5730\u5b9e\u9645\u5546\u4e1a\u9886\u57df\u7684AI Agent\u3002
\n\u4f5c\u4e3a\u4e00\u540d\u7a0b\u5e8f\u5458\/\u5de5\u7a0b\u5e08，\u601d\u8003\u7684\u66f4\u591a\u7684\u53ef\u80fd\u662f\u5982\u4f55\u5c06AI Agent\u843d\u5730\u5b9e\u73b0\u7684\u5c42\u9762\u3002\u5de5\u7a0b\u6846\u67b6\u5f80\u5f80\u662f\u5de5\u7a0b\u5e94\u7528\u5b9e\u73b0\u7684\u57fa\u77f3，\u6846\u67b6\u9009\u578b\u4e5f\u662f\u67b6\u6784\u5e08\u4eec\u8bbe\u8ba1\u4e0e\u5b9e\u73b0\u4e00\u4e2aAI Agent\u667a\u80fd\u4f53\u7684\u9996<\/p>\n","protected":false},"author":2,"featured_media":78820,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[66,50,132,614,958,51,4847],"topic":[],"class_list":["post-78840","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-ai","tag-50","tag-132","tag-614","tag-958","tag-51","tag-4847"],"yoast_head":"\n