[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"pack-detail-agent-deployment-templates-zh":3,"seo:pack:agent-deployment-templates:zh":103},{"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":102},"agent-deployment-templates","🚢","#1F2937","new","本周新建","Agent 上线模板包","10 件套，给真正把 AI agent 推到生产的开发者：FastAPI agent 骨架（Agno、PydanticAI）+ Serverless 目标（Modal、Replicate）+ 沙箱运行时（E2B、Daytona）+ 状态存储和队列（Upstash、LangGraph）+ Kubernetes 部署目标 — 按顺序串起来，agent 才能熬过头 1000 个真实请求。",[16,28,36,46,53,61,69,77,87,94],{"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},299,"f73bc89d-cd16-46bb-af95-3a921a0de059","agno-production-ai-agent-runtime-f73bc89d","Agno — Production AI Agent Runtime","Agno is a runtime for building and managing agentic software at scale. 39.1K+ GitHub stars. Stateful agents, FastAPI serving, 100+ integrations, tracing. Apache 2.0.","Agno",145,0,"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},39,"0313bf39-8bbe-4a50-9445-e5ee8e7280fe","pydanticai-type-safe-ai-agent-framework-0313bf39","PydanticAI — Type-Safe AI Agent Framework","Build production-grade AI agents with type safety, structured outputs, and multi-model support. By the creators of Pydantic and FastAPI.","Pydantic",99,{"id":37,"uuid":38,"slug":39,"title":40,"description":41,"author_name":42,"view_count":43,"vote_count":24,"lang_type":25,"type":44,"type_label":45},2805,"fad6cf5b-d22c-4802-9762-ca47112a05ff","modal-sandboxes-secure-cloud-code-execution-for-ai-agents","Modal Sandboxes — Secure Cloud Code Execution for AI Agents","Modal Sandboxes spin up secure Linux environments for agent-generated code in seconds. Custom images, GPUs, persistent volumes from any Modal Function.","Modal",72,"agent","Agent",{"id":47,"uuid":48,"slug":49,"title":50,"description":51,"author_name":52,"view_count":29,"vote_count":24,"lang_type":25,"type":26,"type_label":27},3112,"154d3162-9681-440c-b4d9-825b073b04a5","modal-examples-serverless-llm-jobs-on-modal","modal-examples — Serverless LLM Jobs on Modal","Learn production patterns for serverless jobs (LLM inference, data pipelines) using Modal’s official examples. Run one and adapt it to your workload.","Script Depot",{"id":54,"uuid":55,"slug":56,"title":57,"description":58,"author_name":59,"view_count":60,"vote_count":24,"lang_type":25,"type":26,"type_label":27},2807,"406d216d-018b-4242-8a26-a4a8df47bb4c","replicate-cog-containerize-ml-models-with-one-yaml-file","Replicate Cog — Containerize ML Models with One YAML File","Cog is Replicate's open-source tool to wrap an ML model in a Docker container. One cog.yaml + predict.py gives you a portable, GPU-aware HTTP model.","Replicate",44,{"id":62,"uuid":63,"slug":64,"title":65,"description":66,"author_name":67,"view_count":68,"vote_count":24,"lang_type":25,"type":26,"type_label":27},3089,"d5ecafac-0501-4f42-adec-5d8b2ac6141a","e2b-secure-sandboxes-for-ai-code","E2B — Secure Sandboxes for AI Code","E2B runs AI-generated code in isolated cloud sandboxes. Install the Python\u002FJS SDK, set `E2B_API_KEY`, then execute commands safely inside a sandbox.","Agent Toolkit",92,{"id":70,"uuid":71,"slug":72,"title":73,"description":74,"author_name":75,"view_count":76,"vote_count":24,"lang_type":25,"type":26,"type_label":27},2813,"3b7e7e34-396e-424e-a2f8-e47decaee4cd","daytona-sdk-programmable-dev-sandboxes-for-ai-agents","Daytona SDK — Programmable Dev Sandboxes for AI Agents","Daytona SDK spawns Linux dev environments in 90 ms. Run agent-generated code, browser automation, ML jobs. Snapshot + fork to branch execution.","Daytona",78,{"id":78,"uuid":79,"slug":80,"title":81,"description":82,"author_name":83,"view_count":84,"vote_count":24,"lang_type":25,"type":85,"type_label":86},691,"e0ed3953-1666-435f-8a4b-f81b4d1447bb","upstash-mcp-serverless-redis-kafka-ai-agents-e0ed3953","Upstash MCP — Serverless Redis & Kafka for AI Agents","MCP server for Upstash serverless Redis and Kafka. Give AI agents access to caching, rate limiting, pub\u002Fsub, and message queues with zero infrastructure. Pay-per-request pricing. 2,000+ stars.","MCP Hub",112,"mcp","MCP",{"id":4,"uuid":88,"slug":89,"title":90,"description":91,"author_name":92,"view_count":93,"vote_count":24,"lang_type":25,"type":26,"type_label":27},"cc1a6ed2-0d82-4379-94f4-15632b4d4967","langgraph-build-stateful-ai-agents-graphs-cc1a6ed2","LangGraph — Build Stateful AI Agents as Graphs","LangChain framework for building resilient, stateful AI agents as graphs. Supports cycles, branching, persistence, human-in-the-loop, and streaming. 28K+ stars.","LangChain",452,{"id":95,"uuid":96,"slug":97,"title":98,"description":99,"author_name":100,"view_count":101,"vote_count":24,"lang_type":25,"type":26,"type_label":27},3150,"3f94c7c7-7f5e-4e7e-8a42-d3c4fd46eaff","agent-sandbox-run-agents-safely-on-kubernetes","Agent Sandbox — Run Agents Safely on Kubernetes","Agent Sandbox provides Kubernetes-first guardrails for agent workloads: resource limits, isolation, and repeatable environments so failures stay contained.","AI Open Source",59,"tokrepo install pack\u002Fagent-deployment-templates",{"pageType":104,"pageKey":8,"locale":105,"title":106,"metaDescription":107,"h1":13,"tldr":108,"bodyMarkdown":109,"faq":110,"schema":126,"internalLinks":132,"citations":145,"wordCount":158,"generatedAt":159},"pack","zh","Agent 上线模板包 — 10 件套带你把 AI agent 推到生产","Agno \u002F PydanticAI \u002F Modal Sandboxes \u002F modal-examples \u002F Replicate Cog \u002F E2B \u002F Daytona \u002F Upstash \u002F LangGraph \u002F Agent Sandbox on Kubernetes — 一套经过深思熟虑的顺序：骨架 → 状态存储 → 沙箱运行时 → 队列 → 部署目标。开源优先，TokRepo 一键安装。","10 件套，带你把 AI agent 从 `python main.py` 跑在笔记本上，走到一个能扛住头 1000 个真实请求的 HTTP 服务。两套 FastAPI agent 骨架（Agno、PydanticAI），两个 Serverless 目标（Modal、Replicate Cog），两个沙箱运行时（E2B、Daytona），一套状态存储 + 队列（Upstash Redis\u002FKafka），一个有状态图框架（LangGraph），一套 Kubernetes 部署模式（Agent Sandbox）。开源优先；只在 SaaS 真值钱的地方点名。","## 这个 pack 包含什么\n\n这是一个真正的工程师在 agent 上线**前那一周**会装的套件 — 不是上线后第一个 OOM 把服务打挂时通宵抢救的那种。每个都是字面意义上的「部署模板」：clone 一个 repo、配几个 env、你就拿到一个能并发处理请求、能持久化状态、能沙箱执行不可信代码、进程挂了能恢复的 agent。开源优先、便宜，每层都能跟下一层串起来。\n\n| # | 工具 | 层 | 干什么 |\n|---|---|---|---|\n| 1 | Agno | agent 骨架（FastAPI） | 生产级 agent runtime，FastAPI 起服务、sessions、集成现成 |\n| 2 | PydanticAI | agent 骨架（类型优先） | 类型安全的 agent 框架 — Pydantic model 当 I\u002FO 契约 |\n| 3 | Modal Sandboxes | serverless 沙箱 | 在隔离的云沙箱里执行 agent 生成的代码 |\n| 4 | modal-examples | serverless 模板 | Modal 上 Serverless LLM 任务的参考 repo |\n| 5 | Replicate Cog | serverless 模板（容器） | 一份 YAML → 一个带 HTTP + Webhook 的容器化模型 |\n| 6 | E2B | 沙箱运行时 | AI 生成代码的安全云沙箱 — Python\u002FJS SDK |\n| 7 | Daytona SDK | 沙箱运行时 | 可编程的开发沙箱 — 支持快照、可重现工作区 |\n| 8 | Upstash（Redis + Kafka） | 状态存储 + 队列 | Serverless Redis 装 session，Kafka 当工作队列 |\n| 9 | LangGraph | 有状态 agent 图 | 把 agent 写成显式状态 + checkpoint 的图 |\n| 10 | Agent Sandbox on Kubernetes | 部署目标 | 在 k8s 上安全跑 agent 的模式 + manifest |\n\n## 推荐安装顺序（骨架 → 状态 → 沙箱 → 队列 → 部署目标）\n\n顺序是有讲究的。**别一上来选部署目标** — 你会被平台特性逼着重写 agent。先在本地把骨架、状态、沙箱跑通，部署目标是最后一个决定。\n\n1. **选一个 agent 骨架**。想要 FastAPI 起服务 + sessions + tracing 都现成的「全套」runtime → **Agno**。想要更小、类型优先、Pydantic model 当 I\u002FO 契约、HTTP 那层自己来 → **PydanticAI**。**选一个**，目标是拿到一个收请求返类型化响应的 `\u002Frun` 端点。先本地跑通。\n2. **加工具之前先加状态存储**。agent 一有 session，你就得找地方装 — 选 **Upstash Redis**（serverless、按请求付费、空闲不收钱）装 session\u002Fcache，**Upstash Kafka**（或任意托管队列）做工作队列（如果单轮可能超 30 秒）。**别把 state 写到本地磁盘**，下次 pod 重启全没了。\n3. **不可信工具调用包到沙箱里**。agent 一旦执行生成的代码、跑 shell、爬网页，就必须隔离。**E2B** 最低门槛 — `from e2b import Sandbox; sbx.run_code(...)` 一行就完事。**Daytona SDK** 是另一种：要长生命周期、可快照的开发工作区时（比如做长任务的 coding agent）用它。**Modal Sandboxes** 是同一种原语但跟 Modal 计算在一起 — 如果你也在 Modal 上部署，用它链路最短。\n4. **请求是脉冲流量选 Serverless 模板**。**Modal**（加 `modal-examples` 当参考 repo）给你一个 Python 装饰器变成 HTTP 端点，带 GPU、缩到零、按秒计费 — 适合请求集中爆发的 agent。**Replicate Cog** 把模型 + handler 打成一个容器，靠 `cog.yaml` 描述；如果你也要同时 serve 一个模型并想合并成一个部署 artifact，用它。\n5. **多步或有状态的流程上 LangGraph**。当 agent 是个多步图（规划 → 搜索 → 反思 → 回答）有分支、有人工介入时，**LangGraph** 给你显式状态 + checkpoint — 单轮崩了能恢复而不是从头来。把它的 checkpointer 绑到第 2 步里的 Redis。\n6. **最后选部署目标**。三条现实路径：**(a) Serverless** — 把 FastAPI app 包进 Modal `@asgi_app`、或 Replicate 的 `cog predict`，作为容器发布。**(b) PaaS** — 用 Dockerfile 推到 Fly\u002FRender\u002FRailway，稳定流量场景最便宜。**(c) Kubernetes** — 需要多租户、gVisor 级隔离，或者已经超出 PaaS 能力时，用 **Agent Sandbox** 当 k8s 跑 agent 的参考模式（每 session 一 pod、每工具调用一沙箱、网络策略默认拒绝）。\n\n## 各部分怎么拼起来\n\n```\n[客户端]\n   │  HTTP \u002Frun\n   ▼\n[FastAPI agent 骨架]   ← Agno 或 PydanticAI\n   │\n   ├─ session\u002Fcache  ──▶  Upstash Redis\n   │\n   ├─ 后台任务       ──▶  Upstash Kafka  ──▶  worker（同镜像）\n   │\n   ├─ 工具: run_code ──▶  E2B \u002F Daytona \u002F Modal Sandbox\n   │\n   ├─ 图状态         ──▶  LangGraph checkpointer (Redis)\n   │\n   ▼\n[部署目标]  ── Modal @asgi_app  \u002F  Replicate Cog  \u002F  k8s (Agent Sandbox)\n```\n\n四件套组合 **骨架 + 状态存储 + 沙箱 + 部署目标** 是「最小可行生产 agent」。少哪一个都顶不过一周：没状态 → 用户骂失忆；没沙箱 → 幻觉调出的 `rm -rf` 干掉你的服务；没骨架 → 你又写了一遍 FastAPI 中间件还写得不如别人；没部署目标 → 你根本接不到流量。\n\n## 你会遇到的取舍\n\n- **Agno vs PydanticAI** — Agno 是更大的框架，sessions \u002F FastAPI app \u002F 集成 \u002F tracing 现成；代价是你得忍受它的主张。PydanticAI 更小更类型优先，HTTP 那层你自己拼。两周要上线的团队 → Agno。已经有 FastAPI 约定的团队 → PydanticAI。\n- **E2B vs Daytona vs Modal Sandbox** — E2B 是最低门槛的临时代码执行沙箱（Python SDK，默认安全）。Daytona 长项是需要持久化、可快照的工作区（长生命周期的 coding agent）。Modal Sandbox 是「你已经在 Modal」的最优解 — 同 auth、同账单、跟模型调用同机房延迟最低。\n- **Modal vs Replicate Cog vs k8s** — Modal 缩到零、按秒计费、Python 即部署单元；脉冲流量最划算。Replicate Cog 是一个容器 + `cog.yaml`；同时 serve 模型 + agent 时合并部署最简单。Kubernetes（走 Agent Sandbox 模式）适合真需要多租户、gVisor 隔离、或者已经超出托管平台能力。\n- **LangGraph vs 手写状态机** — 单轮 agent（一次 LLM 调 + 工具）别拉 LangGraph 进来，纯负担。多轮图 + 分支 + 重试 + 人工介入，LangGraph 的 checkpointer 让崩溃可恢复，值这个体积。\n- **Upstash Redis vs 自托管 Redis** — Upstash 是 serverless、按请求计费；超过 ~1000 万 commands\u002F月 之后 $20 的 Redis VM 反而更便宜。迁移就是改个 URL。**别提前优化**。\n\n## 常见踩坑\n\n- **session state 写本地磁盘或进程内存**。下次 pod 重启全没了，用户骂你失忆。**第一天就把 state 写进 Redis（或你的 DB）**，不是出事之后再改。\n- **工具调用没沙箱**。agent 第一次幻觉出 `subprocess.run(['rm', '-rf', '\u002F'])` 而你的服务真跑了，生产集群就完了。**只要工具包含 shell 或代码执行，E2B\u002FDaytona\u002FModal Sandbox 不是可选项**。\n- **Serverless 跑长 agent 轮次**。大多数 serverless 有最大执行时间（Lambda 15min、Vercel 5min、Modal 最长 24h）。如果单轮可能 30+ 分钟，要么选 Modal（长超时），要么把活推进队列让 HTTP 请求立刻返个 job ID。\n- **FastAPI 骨架里没设请求级超时**。没超时一次卡死的 LLM 调用就能把 worker 池耗干。**HTTP 边界、LLM 客户端、工具调用三层都要显式设超时**。\n- **把完整 prompt + response 全打日志**。开发期感觉很爽，生产环境 PII 漏进不合规的日志聚合器。**截断、脱敏、采样后再打**，完整 trace 留给 LLM observability（Langfuse、Phoenix）走访问控制。",[111,114,117,120,123],{"q":112,"a":113},"10 个工具真的都需要吗？看着好多。","你需要每**层**选一个，不是 10 个全装。pack 在同层列了备选（2 个骨架、3 个沙箱、3 条部署路径）— 按你的规模挑。1 人独立开发者最小可行生产 agent：Agno（骨架）+ Upstash Redis（状态）+ E2B（沙箱）+ Modal（部署）— 4 件套，一下午就上线。agent 长成多步图再加 LangGraph。撑不下托管平台再上 Agent Sandbox on Kubernetes。",{"q":115,"a":116},"这套月成本大概多少？","小 agent 每天处理几千请求场景：Modal ~$5-50\u002F月（按秒计费、缩到零），Upstash Redis 免费层或 ~$10\u002F月，E2B 免费层（100h\u002F月）或稳定用 ~$30\u002F月，开源骨架和 LangGraph 不要钱。合计 $15-100\u002F月端到端。**真正的成本是 LLM 调用**，通常远超基础设施；这套挑法是让基础设施在 model 账单面前是个零头。",{"q":118,"a":119},"这个跟 LLM Observability pack 重叠吗？","不同层。这个 pack 是**部署层** — agent 进程怎么活、怎么持久化、怎么 serve 流量。LLM Observability（Langfuse \u002F Phoenix \u002F AgentOps）是**应用语义层** — prompt \u002F trace \u002F eval。**两个都要**。骨架第一天就发 OpenTelemetry；observability 那套接住。大多数团队先上这个 pack（agent 还没部署观测什么），同周再上 observability。",{"q":121,"a":122},"如果我已经在 Modal 上跑了，为啥还推 E2B？","如果你的计算已经在 Modal 上，**Modal Sandboxes 就是答案** — 同 auth、同账单、模型调用低延迟、不引入第二朵云。E2B 的赢面是你部署在**别的目标**上（Fly \u002F Replicate \u002F k8s）但需要一个不拖第二个账号进来的沙箱。Daytona 的赢面是沙箱要活几小时甚至几天（coding agent 的持久化工作区），而不是几秒。",{"q":124,"a":125},"这套能跑长任务、多步的研究型 agent 吗（不是聊天 agent）？","能 — 这套套件本来就是奔着这种场景设计的。用 LangGraph 写图 + checkpoint（单轮崩了能恢复），checkpointer state 放 Upstash Redis，每个长步骤推进 Upstash Kafka 跑在 worker pod 里，任何执行代码的工具用 E2B 或 Daytona 沙箱，部署在 Modal（长超时）或走 Agent Sandbox 模式上 Kubernetes（需要真多租户）。HTTP `\u002Frun` 立刻返 job ID，客户端 poll 或订阅拿结果。",{"@context":127,"@type":128,"name":13,"description":129,"numberOfItems":130,"inLanguage":131},"https:\u002F\u002Fschema.org","ItemList","10 个开源优先的工具，带开发者把 AI agent 从笔记本搬到能 serve 流量的 HTTP 端点：骨架 \u002F 状态存储 \u002F 沙箱运行时 \u002F 队列 \u002F 部署目标。",10,"zh-CN",[133,137,141],{"url":134,"anchor":135,"reason":136},"\u002Fzh\u002Fpacks\u002Fdeploy-monitor-observability","上线 + 监控 + 可观测性 一站套件","agent 上线之后这个 pack 负责把部署 → 追踪 → 日志 → 告警串起来",{"url":138,"anchor":139,"reason":140},"\u002Fzh\u002Fpacks\u002Fagent-memory-layer","Agent 记忆层 pack","session 状态之后那一层长期记忆 — 向量库、语义记忆、检索",{"url":142,"anchor":143,"reason":144},"\u002Fzh\u002Fai-tools-for\u002Fdevops","DevOps 工具集（面向 AI Agent）","更全的部署目标、容器工具、运行时模式目录",[146,150,154],{"claim":147,"source_name":148,"source_url":149},"E2B 提供执行 AI 生成代码的安全云沙箱","E2B 官方文档","https:\u002F\u002Fe2b.dev\u002Fdocs",{"claim":151,"source_name":152,"source_url":153},"Modal Sandboxes 允许在隔离的云容器中运行不可信代码","Modal Sandboxes 文档","https:\u002F\u002Fmodal.com\u002Fdocs\u002Fguide\u002Fsandbox",{"claim":155,"source_name":156,"source_url":157},"LangGraph 把 agent 构建成有状态、带 checkpoint 的图","LangGraph 官方文档","https:\u002F\u002Flangchain-ai.github.io\u002Flanggraph\u002F",905,"2026-05-22T12:00:00Z"]