[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"pack-detail-agent-frameworks-multilang-zh":3,"seo:pack:agent-frameworks-multilang:zh":68},{"code":4,"message":5,"data":6},200,"操作成功",{"pack":7},{"slug":8,"icon":9,"tone":10,"status":11,"status_label":12,"title":13,"description":14,"items":15,"install_cmd":67},"agent-frameworks-multilang","🌐","#0891B2","stable","稳定","多语言 Agent 框架","Spring AI（Java）\u002F LangChain4j \u002F Candle（Rust）\u002F Mastra（TS）\u002F FastHTML（Python）\u002F Axum（Rust API）— 每种栈都有 agent 框架。",[16,28,36,44,51,59],{"id":17,"uuid":18,"slug":19,"title":20,"description":21,"author_name":22,"view_count":23,"vote_count":24,"lang_type":25,"type":26,"type_label":27},326,"98ae1961-8a0b-456f-9ff2-10852b832001","spring-ai-ai-engineering-java-spring-98ae1961","Spring AI — AI Engineering for Java\u002FSpring","Spring AI provides Spring-friendly APIs for AI apps. 8.4K+ stars. Chat, embeddings, RAG, vector DBs, function calling. Major providers. Apache 2.0.","Skill Factory",294,0,"en","skill","Skill",{"id":29,"uuid":30,"slug":31,"title":32,"description":33,"author_name":34,"view_count":35,"vote_count":24,"lang_type":25,"type":26,"type_label":27},327,"f7069ed9-81bd-4e67-a727-44e37eb529e7","langchain4j-llm-integration-java-f7069ed9","LangChain4j — LLM Integration for Java","LangChain4j integrates 20+ LLM providers and 30+ vector stores into Java apps. 11.4K+ stars. Unified API, RAG, MCP, Spring Boot. Apache 2.0.","LangChain",345,{"id":37,"uuid":38,"slug":39,"title":40,"description":41,"author_name":42,"view_count":43,"vote_count":24,"lang_type":25,"type":26,"type_label":27},1302,"b113c394-37db-11f1-9bc6-00163e2b0d79","candle-minimalist-machine-learning-framework-rust-b113c394","Candle — Minimalist Machine Learning Framework for Rust","Candle is a Rust-native ML framework focused on inference performance, small binaries, and serverless deployment. It runs Llama, Whisper, Stable Diffusion, and other PyTorch models in pure Rust — no Python required.","AI Open Source",376,{"id":45,"uuid":46,"slug":47,"title":48,"description":49,"author_name":22,"view_count":50,"vote_count":24,"lang_type":25,"type":26,"type_label":27},739,"143825f7-ee67-49a1-a994-6908bb1df20f","fasthtml-build-ai-web-apps-pure-python-143825f7","FastHTML — Build AI Web Apps in Pure Python","Modern Python web framework that generates HTML from Python functions. No JavaScript, no templates. Perfect for building AI tool dashboards and agent UIs rapidly.",205,{"id":52,"uuid":53,"slug":54,"title":55,"description":56,"author_name":57,"view_count":58,"vote_count":24,"lang_type":25,"type":26,"type_label":27},633,"3e118616-e727-4dc7-a561-db39e91cadcd","mastra-typescript-ai-agent-framework-3e118616","Mastra — TypeScript AI Agent Toolkit","Production TypeScript framework for building AI agents with tool use, workflows, RAG, and memory. First-class MCP support. Deploy anywhere Node.js runs. 9,000+ GitHub stars.","Mastra",325,{"id":60,"uuid":61,"slug":62,"title":63,"description":64,"author_name":65,"view_count":66,"vote_count":24,"lang_type":25,"type":26,"type_label":27},1088,"412ff341-3634-11f1-9bc6-00163e2b0d79","axum-ergonomic-modular-web-framework-rust-412ff341","Axum — Ergonomic Modular Web Framework for Rust","Axum is a web application framework built on Tokio, Tower, and Hyper. Focuses on ergonomics and modularity with a macro-free routing API, seamless Tower middleware integration, and type-safe extractors. The official Tokio team web framework.","Script Depot",328,"tokrepo install pack\u002Fagent-frameworks-multilang",{"pageType":69,"pageKey":8,"locale":70,"title":71,"metaDescription":72,"h1":13,"tldr":73,"bodyMarkdown":74,"faq":75,"schema":91,"internalLinks":101,"citations":114,"wordCount":127,"generatedAt":128},"pack","zh","多语言 Agent 框架：Java \u002F Rust \u002F TS \u002F Python 全覆盖","Spring AI（Java）\u002F LangChain4j \u002F Candle（Rust）\u002F Mastra（TS）\u002F FastHTML（Python）\u002F Axum（Rust API）— 每种栈都有 agent 框架。TokRepo 一条命令装齐。","六个框架让 agent 应用在 Java \u002F Rust \u002F TypeScript \u002F Python 都成为一等公民 —— 选匹配你现有栈的，不用为了 LLM 功能再嫁接一个 Python 服务。TokRepo CLI 一条命令装齐。","## 这个 pack 装了什么\n\n这个包收齐了 **六个 agent 框架**，让团队留在自己的语言生态里，不用为了 LLM 功能在 Java 或 Rust 后端旁边硬塞一个 Python 服务。覆盖 Python 之外的四种语言：Java \u002F Rust \u002F TypeScript \u002F Python（FastHTML 做全栈渲染）。\n\n| # | 框架 | 语言 | 适合 |\n|---|---|---|---|\n| 1 | Spring AI | Java | Spring Boot 应用加 LLM 功能 |\n| 2 | LangChain4j | Java\u002FKotlin | LangChain 在 JVM 的社区移植 |\n| 3 | Candle | Rust | 本地模型推理 |\n| 4 | Mastra | TypeScript | 全栈 TS agent + 工作流框架 |\n| 5 | FastHTML | Python | 服务端渲染的 agent UI |\n| 6 | Axum | Rust | 高吞吐 agent API |\n\n这个 pack 反映了一个现实：LLM 功能现在出现在每种后端里，不只 Python。Spring Boot 的电商后端不想为了聊天功能再起一个 FastAPI 服务 —— Spring AI 跑在同一个 JVM 里。Rust API 网关不想要 Python sidecar —— Axum + Candle 编译成一个二进制。\n\n## 为什么语言原生很重要\n\nPython 优先的 agent 世界（LangChain \u002F LangGraph \u002F CrewAI）当 Python 已经是主栈时很好。如果你是 Java 公司，没这个 pack 你只有三个糟糕选项：\n1. 跑 Python sidecar 服务（额外部署 \u002F 独立可观测性 \u002F 两个进程之间 marshalling JSON）\n2. 用 SaaS LLM 网关（厂商锁定 \u002F 延迟 \u002F 费用）\n3. 在 Spring controller 里手写 OpenAI HTTP 调用（没有抽象 \u002F 没有评估框架 \u002F 没有 agent 循环）\n\nSpring AI 和 LangChain4j 修这个。它们给 Java 提供 LangChain 给 Python 的同款原语 —— chat 模型 \u002F 工具调用 \u002F prompt 模板 \u002F 向量存储 \u002F agent 循环 —— 用地道的 Spring 或 Kotlin DSL。Rust（Candle 推理 + Axum API 面）和 TypeScript（Mastra 框架）同理。\n\n代价是功能滞后。LangChain 第一天就发新模型集成；Spring AI 要几周；LangChain4j 有时比 Spring AI 还快因为是社区驱动。「模型昨天发的，只有 Python 接了」是真实场景。\n\n## 一条命令装齐\n\n```bash\n# 装整个 pack，把每个框架的 starter 项目放进\ntokrepo install pack\u002Fagent-frameworks-multilang\n\n# 或按语言装\ntokrepo install spring-ai\ntokrepo install langchain4j\ntokrepo install mastra\ntokrepo install candle\n```\n\nTokRepo CLI 处理每个生态的包管理器 —— Spring AI 和 LangChain4j 用 Maven，Candle 和 Axum 用 Cargo，Mastra 用 npm\u002Fpnpm，FastHTML 用 uv。每个 starter 项目都带一个会调真模型的通过测试，写业务逻辑前先验证 API key。\n\n## 几个常见坑\n\n- **Spring AI vs LangChain4j 不是「官方 vs 非官方」**。Spring AI 来自 VMware\u002FBroadcom；LangChain4j 是社区。两个都广泛使用。按你用的模型的功能对齐和人体工学偏好（Spring DSL vs LangChain 语义）选\n- **Mastra 还年轻**。维护得好但小版本之间 API 有变化。生产钉具体版本，升级前看 release notes。生产用户一般等两个小版本再升\n- **Candle 做推理，不做编排**。它在本地跑模型（Llama \u002F Mistral 等）给你 embedding —— 但不是完整 agent 框架。要 Rust-only 栈把 Axum 配上做 API 层\n- **FastHTML 基于 HTMX**，不是 SPA 框架。团队预期 React 或 Vue 的话 FastHTML 会很别扭。适合想要服务端渲染的 Python UI 直接接 agent，不要单独前端 repo 的场景\n- **工具调用格式各家不同**。OpenAI \u002F Anthropic \u002F Gemini 的工具调用 JSON schema 都不一样。每个框架抽象方式不同 —— Spring AI 的 `@Tool` 注解 \u002F LangChain4j 的反射 `@Tool` \u002F Mastra 的类型化 `createTool()`。别假设工具定义能跨框架移植\n\n## 单这个 pack 不够时\n\n这个 pack 给你你语言里的 runtime。你还需要：\n- **LLM 提供商**：OpenAI \u002F Anthropic \u002F Bedrock \u002F Ollama 自托管。六个框架都支持配置切多 provider\n- **向量存储**（如果做 RAG）：见向量数据库横评 pack —— 多数条目都有 Java \u002F Rust \u002F TS 原生客户端\n- **评估管道**：LLM 评测 & 护栏 pack 里的 Promptfoo 是语言无关的 —— 不管后端啥语言都能指你的端点\n\n纯 Python 栈、不关心多语言的，去看 [Python Agent 框架](\u002Fzh\u002Fpacks\u002Fpython-agent-frameworks)。不挑语言、想平台无关并列对比，去看 [多 agent 框架](\u002Fzh\u002Fpacks\u002Fmulti-agent-frameworks)。",[76,79,82,85,88],{"q":77,"a":78},"这些框架免费吗？","六个全是 Apache 2.0 或 MIT 开源 —— 框架本身不收按席位费。LLM API 调用（OpenAI \u002F Anthropic 等）该付的还是要付，跟语言无关。Spring AI 来自 Broadcom 但免费；LangChain4j 社区驱动免费；Mastra 有 VC，开源免费 + 可选云服务。Candle \u002F Axum \u002F FastHTML 是纯 OSS。",{"q":80,"a":81},"Spring AI 跟 LangChain4j 怎么比？","Spring AI 是 Spring 团队做的标准答案 —— 跟 Spring Boot autoconfig \u002F Spring Data \u002F Spring Security 紧密集成。LangChain4j 社区驱动，把 LangChain 概念模型（chain \u002F agent \u002F callback）移植到 Java，功能覆盖更广。Spring 公司想要标准用法选 Spring AI；想要 LangChain 心智模型在 JVM 里、要最大集成广度选 LangChain4j。",{"q":83,"a":84},"Claude Code 或 Cursor 能用吗？","这些是运行时框架，不是编辑器集成。框架本身跑在你的服务里（Spring Boot \u002F Axum 二进制 \u002F Mastra 应用）。你的编辑器（Claude Code \u002F Cursor）帮你写框架代码。Cursor 和 Claude Code 对 Java \u002F Rust \u002F TS 都支持得不错，Spring AI 还能从 JetBrains 的 Spring IDE 插件得到很好的自动补全。",{"q":86,"a":87},"和 Python Agent 框架 pack 的区别？","Python Agent 框架是纯 Python —— 五个框架都跑 CPython。这个 pack 是反面：暴露*非 Python*选项。已经定 Python 了那个 pack 更密集；在 Java\u002FRust\u002FTS 上或还在挑语言这个 pack 是起点。两者互补不重叠。",{"q":89,"a":90},"Mastra 的运维坑是什么？","Mastra 有强意见认为工作流跟 agent 应该住同一个项目，共享类型。代码库干净是好的，但工作流引擎和 agent 框架版本紧耦合。Mastra 升级一个工作流坏了，你不能只回滚 agent 不回滚工作流。Mastra 版本升级要专门测试阶段。",{"@context":92,"@type":93,"name":94,"description":95,"numberOfItems":96,"publisher":97},"https:\u002F\u002Fschema.org","CollectionPage","Agent Frameworks Across Languages","Spring AI for Java, LangChain4j, Candle for Rust, Mastra for TypeScript, FastHTML for Python, Axum for Rust APIs — agent frameworks beyond Python.",6,{"@type":98,"name":99,"url":100},"Organization","TokRepo","https:\u002F\u002Ftokrepo.com",[102,106,110],{"url":103,"anchor":104,"reason":105},"\u002Fzh\u002Fpacks\u002Fpython-agent-frameworks","Python Agent 框架","纯 Python 的对应 pack",{"url":107,"anchor":108,"reason":109},"\u002Fzh\u002Fpacks\u002Fmulti-agent-frameworks","多 agent 框架","平台无关的多 agent 比较",{"url":111,"anchor":112,"reason":113},"\u002Fzh\u002Fpacks\u002Fpostgres-for-agents","Postgres for Agent","每个栈都会用到的 SQL 接口",[115,119,123],{"claim":116,"source_name":117,"source_url":118},"Spring AI official Java framework for AI integration","spring-projects\u002Fspring-ai","https:\u002F\u002Fgithub.com\u002Fspring-projects\u002Fspring-ai",{"claim":120,"source_name":121,"source_url":122},"LangChain4j community port to JVM","langchain4j\u002Flangchain4j","https:\u002F\u002Fgithub.com\u002Flangchain4j\u002Flangchain4j",{"claim":124,"source_name":125,"source_url":126},"Mastra TypeScript-first agent framework","mastra-ai\u002Fmastra","https:\u002F\u002Fgithub.com\u002Fmastra-ai\u002Fmastra",593,"2026-05-02T15:00:00Z"]