Agent = LLM + 上下文 + 工具——本书围绕这个核心公式,用 10 章把 AI Agent 从原理讲到工程实战。全书正文、配图、92 个配套实验全部开源,欢迎亲手把实验跑一遍。
中文正文源码位于 book/,配套实验代码在 chapter1/ ~ chapter10/。
全书围绕核心公式 Agent = LLM + 上下文 + 工具 展开,十章层层递进:
| 章 | 主题 | 一句话核心 | 正文 | 代码 |
|---|---|---|---|---|
| 1 | 🚀 Agent 基础知识 | Agent = LLM + 上下文 + 工具;Harness 工程才是竞争力 | 读 | 4 |
| 2 | 🎯 上下文工程 | 上下文决定能力上限:KV Cache、提示工程、Agent Skills、上下文压缩 | 读 | 9 |
| 3 | 📚 用户记忆和知识库 | 跨会话记住用户、接入外部知识:用户记忆、RAG、结构化索引、知识图谱 | 读 | 13 |
| 4 | 🛠️ 工具 | 工具是 Agent 的双手:MCP 协议、感知/执行/协作三类工具、事件驱动异步 Agent、主动工具发现 | 读 | 7 |
| 5 | 💻 Coding Agent 与代码生成 | 代码是「能创造新工具的工具」,生产级 Coding Agent 全景 | 读 | 12 |
| 6 | 🎯 Agent 的评估 | 把表现变成可比较信号:评估环境、指标、统计显著性、评估驱动选型 | 读 | 11 |
| 7 | 🧠 模型后训练 | 预训练/SFT/RL 三阶段:何时选 SFT、何时选 RL,工具调用内化、样本效率 | 读 | 16 |
| 8 | 🔄 Agent 的自我进化 | 不改权重也能成长:经验学习、从工具使用者到创造者 | 读 | 6 |
| 9 | 🎙️ 多模态与实时交互 | 从文本扩展到语音、GUI、物理世界:语音三范式、Computer Use、机器人 | 读 | 7 |
| 10 | 🤝 多 Agent 协作 | 群体智能高于个体:协作框架、上下文共享/隔离、涌现的「Agent 社会」 | 读 | 7 |
💡 读 = 在 GitHub 网页直接读章节正文(markdown);N = 该章配套项目数,点击查看代码。项目类型说明(✅ 可运行 / 📖 复现 / 🚧 设计)见各章 README。
📚 如何高效阅读本书?详见 学习建议(核心理念、学习路径、难度分级、实践建议)。
建议申请下面几个平台的 API Key 方便学习。模型选型可参考 这篇指南。
| 平台 | 链接 | 特色 | 访问节点 |
|---|---|---|---|
| Kimi(月之暗面) | https://platform.moonshot.cn/ | Kimi 系列,Coding、Agent 能力强 | 中国大陆 |
| 智谱 GLM | https://open.bigmodel.cn/ | GLM-5.2 等,Coding、Agent 能力强 | 中国大陆 |
| Siliconflow | https://siliconflow.cn/ | 各种开源模型(DeepSeek、Qwen 等),中国大陆访问速度快 | 中国大陆 |
| DeepSeek | https://platform.deepseek.com/ | DeepSeek 官方 API | 全球 + 中国大陆 |
| Krill AI | www.krill-ai.com | 一站式访问全球及国内主流模型(OpenAI、Claude、Gemini、Grok、Kimi、GLM、DeepSeek、Qwen、Minimax) | 全球 + 中国大陆 |
| OpenRouter | https://openrouter.ai/ | 一站式访问全球及国内主流模型(GPT、Claude、Gemini、Kimi、GLM、DeepSeek、Qwen 等) | 全球 |
第 6、7、9、10 章的评测基准、训练框架、机器人平台等 19 个外部仓库未内置(出于体积与版权),需要自行克隆到对应目录。
🔧 展开克隆命令(共 19 个外部仓库)
# 第 6 章 · 评测基准
git clone https://github.com/google-research/android_world.git chapter6/android_world
git clone https://huggingface.co/datasets/gaia-benchmark/GAIA chapter6/GAIA
git clone https://github.com/xlang-ai/OSWorld.git chapter6/OSWorld
git clone https://github.com/SWE-bench/SWE-bench.git chapter6/SWE-bench
git clone https://github.com/sierra-research/tau2-bench.git chapter6/tau2-bench
git clone https://github.com/laude-institute/terminal-bench.git chapter6/terminal-bench
# 第 7 章 · 训练框架(bojieli/* 为本书适配的分支)
git clone https://github.com/bojieli/minimind.git chapter7/MiniMind-pretrain/minimind # 实验 7-3 从零训 LLM
git clone https://github.com/bojieli/minimind-v.git chapter7/MiniMind-pretrain/minimind-v # 实验 7-4 从零训 VLM(投影层)
git clone https://github.com/bojieli/AdaptThink.git chapter7/AdaptThink-original
git clone https://github.com/bojieli/AWorld.git chapter7/AWorld
git clone https://github.com/bojieli/SFTvsRL.git chapter7/SFTvsRL
git clone https://github.com/bojieli/verl.git chapter7/verl
git clone https://github.com/thinking-machines-lab/tinker-cookbook.git chapter7/tinker-cookbook
git clone https://github.com/19PINE-AI/rlvp.git chapter7/RLVP/rlvp # 实验 7-14 RLVP 论文代码
git clone https://github.com/PRIME-RL/SimpleVLA-RL.git chapter7/SimpleVLA-RL/SimpleVLA-RL # 实验 7-13 视觉-语言-动作 RL
# 第 9 章 · 浏览器自动化与 Claude 示例
git clone https://github.com/browser-use/browser-use.git chapter9/browser-use
git clone https://github.com/anthropics/claude-quickstarts.git chapter9/claude-quickstarts
# 第 10 章 · 双 Agent 架构(已独立为 TalkAct 项目)+ 斯坦福 AI 小镇
git clone https://github.com/19PINE-AI/TalkAct.git chapter10/use-computer-while-calling
git clone https://github.com/joonspk-research/generative_agents.git chapter10/generative_agents # 实验 10-7 斯坦福 AI 小镇各项目 README 如标注了特定 commit,请按说明
git checkout到对应版本以保证复现一致。第 10 章use-computer-while-calling已发展为独立维护的 19PINE-AI/TalkAct,本仓库不内置该目录,用上面的克隆命令获取。
下面这些实验无专属 clone 命令,但有特定的复现方式:
| 实验 | 类型 | 说明 |
|---|---|---|
| 6-2 / 6-3 / 6-4 / 6-9 | 📝 读者练习 | 人肉基准、记忆评估、JSON Cards vs RAG、记忆选型——改造复用第 3 章 user-memory / user-memory-evaluation / contextual-retrieval |
| 5-12 | 📝 读者练习 | 能创造 Agent 的 Agent——基于 chapter5/coding-agent 自举扩展 |
| 7-8 | 📝 读者练习 | Prompt 蒸馏——落地实现见 chapter8/prompt-distillation(跨章复用) |
| 7-9 | 📝 读者练习 | CoT 蒸馏 [扩展]——配套实现见 chapter7/cot-distillation(含 SFT 数据生成与规则验证器) |
| 6-11 | 🤖 仿真评估 | OpenVLA + RoboTwin2——VLA 训练/环境依赖见 chapter7/SimpleVLA-RL 的 README |
| 9-8 / 9-9 | 🔧 真实硬件 | XLeRobot 遥操作与 LLM Agent 控制——需 SO-100 机械臂,Teleop · LLM Agent |
| 9-10 | 🔧 真实硬件 | RGB 零样本 Sim2Real 抓取——StoneT2000/lerobot-sim2real(仿真可纯 GPU,部署需 SO-100) |