LangChain 7月更新：NemoClaw Deep Agents 蓝图、OpenWiki Brains 及更多新功能

本月，Jensen Huang 和 Harrison 讨论了为什么公司需要开放代理系统。此外还有新内容：我们的 NVIDIA 蓝图以及免费 LangSmith Sandboxes 试用。

另外，Interrupt 秋季将巡回至纽约和伦敦——在门票售罄前获取您的门票：https://interrupt.langchain.com/。

Jensen 和 Harrison 讨论了 AI 的未来将建立在开放代理系统之上。观看视频，了解 NVIDIA NemoClaw for LangChain 深度代理蓝图如何指向一个团队可以拥有、调优和控制代理系统的未来。

开始使用 NemoClaw + 深度代理代码 →：https://www.langchain.com/blog/deep-agents-code-on-nemoclaw-a-governed-blueprint-for-your-most-sensitive-code

在旧金山 Interrupt 汇聚代理构建者社区后，我们将会议带到纽约和伦敦。聆听行业领导者的分享，包括 Harrison Chase，以及来自 JPMorgan Chase、Apollo 和 AstraZeneca 的演讲嘉宾，并参加关于构建、调试和发布可靠代理的实操工作坊。

名额有限——今天注册 →：https://interrupt.langchain.com/

📦 LangSmith Sandboxes 现提供免费试用，让您的代理在安全、隔离的环境中运行代码、测试更改，并安全地完成长时间运行的工作。阅读 Sandboxes 文档 →：https://docs.langchain.com/langsmith/sandboxes‍

💬 LangSmith Fleet：一键将代理带入 Slack。在频道中标记代理，在线程中处理跟进和 HITL 审批，发送和接收文档，等等。阅读公告 →：https://www.langchain.com/blog/new-in-langsmith-fleet-bring-agents-into-slack-in-one-click

🗣️ LangSmith 中的语音代理跟踪：在单个跟踪中捕获通过 Pipecat、LiveKit、OpenAI Realtime 和 Gemini Live 构建的代理的音频、STT 和 TTS 延迟、中断、工具调用等。了解更多 →：https://docs.langchain.com/langsmith/integrations

🧠 OpenWiki Brains：通用记忆层，将 Gmail、Notion、git 仓库、X、Hacker News 和网页搜索等来源转变为本地 Wiki，供您的代理使用并自动保持更新。阅读博客 →：https://www.langchain.com/blog/introducing-openwiki-brains-general-purpose-wiki-memory-for-agents

🔎 Deep Agents 和 Harbor 现在提供一个统一的评估代理堆栈，包括隔离的沙箱运行、并行试验以及展示每个代理成功或失败原因的 LangSmith 跟踪。了解更多 →：https://www.langchain.com/blog/unified-stack-for-evaluating-agents

🧩 Deep Agents 中的 RLMs：使用动态子代理将大上下文任务分解为程序化子代理调用，帮助代理减少上下文衰减，并在大输入上协调工作。阅读指南 →：https://www.langchain.com/blog/how-to-use-rlms-in-deep-agents

学习如何使用 Deep Agents（LangChain 的开源代理框架）构建复杂工作流的长期运行代理。您将探索代理框架是什么，它如何加速开发，以及如何使用 LangSmith 改进您的代理。

报名课程 →：https://academy.langchain.com/courses/foundation-introduction-to-deepagents

🗽 纽约 · 7月24日 AWS AI Builder Lab: 代理对决比赛 ‍ RSVP →：https://www.aicamp.ai/event/eventdetails/W2026072406

🔁 波士顿 · 7月27日 LangChain Meetup: 循环工程与开放模型 ‍ RSVP →：https://luma.com/iaie4h2x

🧠 旧金山 · 7月28日 LangChain Meetup: 使用 Cognition 的 LLM 维基 ‍ RSVP →：https://luma.com/mylwoabu

🏎️ 拉斯维加斯 · 8月4日 高风险高速度与 Elastic & Google Cloud ‍ RSVP →：https://events.elastic.co/f1arcadelasvegas/

🤖 亚特兰大 · 硅谷 · 波士顿 · 多个日期 LangSmith 巡回演示：用代理构建代理 参加半天的技术会议、现场演示、工程领导者聚焦以及专注于在生产中构建和改进代理的实践实验室。 ‍查找您的城市 →：https://events.langchain.com/LangSmithRoadshow/

施耐德电气利用 LangSmith 为 60 多款 AI 产品构建了企业级 LLMOps 基础，向 200 多名工程师和主题专家提供了共享的可观察性、评估和部署标准。阅读故事 →：https://www.langchain.com/blog/how-schneider-electric-built-their-llmops-foundations-at-enterprise-scale-with-langsmith

Pendo 使用 LangSmith 推出了 Novus，这是一款产品代理，它能够将用户行为和会话回放转化为可用的代码修复，并在 PM 审核的评估中达到 90% 以上的成功率。阅读完整故事 →：https://www.langchain.com/blog/how-pendo-used-langsmith-to-trace-novus-from-user-behavior-to-code-fixes

如有任何其他问题或反馈，请发送电子邮件至 hello@langchain.dev。

LangSmith，我们的代理工程平台，帮助开发者调试每一个代理决策、评估更改，并一键部署。

This month, Jensen Huang and Harrison talk why companies need open agent systems. Also new: our NVIDIA blueprint and a free LangSmith Sandboxes trial.

Plus, Interrupt is hitting the road to NYC and London this fall — get your tickets：https://interrupt.langchain.com/ before they're gone.

Jensen and Harrison talk about why the future of AI will be built on open agent systems. Watch to learn how the NVIDIA NemoClaw for LangChain Deep Agents blueprint points toward a future where teams build agent systems they can own, tune, and control.

Get started with NemoClaw + Deep Agents Code →：https://www.langchain.com/blog/deep-agents-code-on-nemoclaw-a-governed-blueprint-for-your-most-sensitive-code

After bringing together the agent builder community at Interrupt in San Francisco, we're taking the conference on the road to in NYC and London. Hear from industry leaders, including Harrison Chase, with speakers from JPMorganChase, Apollo, and AstraZeneca, and join hands-on workshops on building, debugging, and shipping reliable agents.

Spots are limited - register today →：https://interrupt.langchain.com/

📦 LangSmith Sandboxes now offers a free trial, so you can give your agents as secure, isolated environment to run code, test changes, and complete long-running work safely. Read Sandboxes docs →：https://docs.langchain.com/langsmith/sandboxes‍

💬 LangSmith Fleet: Bring agents into Slack with one click. Tag agents in channels, handle follow-ups and HITL approvals in threads, send and receive documents, and more. Read announcement →：https://www.langchain.com/blog/new-in-langsmith-fleet-bring-agents-into-slack-in-one-click

🗣️ Voice agent tracing in LangSmith: Capture audio, STT and TTS latency, interruptions, tool calls, and more in one trace for agents built with Pipecat, LiveKit, OpenAI Realtime, and Gemini Live. Learn more →：https://docs.langchain.com/langsmith/integrations

🧠 OpenWiki Brains: A general-purpose memory layer that turns sources like Gmail, Notion, git repos, X, Hacker News, and web search into a local wiki your agents can use and keep updated automatically. Read the blog →：https://www.langchain.com/blog/introducing-openwiki-brains-general-purpose-wiki-memory-for-agents

🔎 Deep Agents and Harbor now provide a unified stack for evaluating agents, with isolated sandbox runs, parallel trials, and LangSmith traces that show why each agent succeeds or fails. Learn more →：https://www.langchain.com/blog/unified-stack-for-evaluating-agents

🧩 RLMs in Deep Agents: Use dynamic subagents to break large-context tasks into programmatic subagent calls, helping agents reduce context rot and orchestrate work over large inputs. Read guide →：https://www.langchain.com/blog/how-to-use-rlms-in-deep-agents

Learn how to build long-running agents for complex workflows with Deep Agents, LangChain's open-source agent harness. You'll explore what an agent harness is, how it accelerates development, and how to use LangSmith to improve your agents.

Enroll in the course →：https://academy.langchain.com/courses/foundation-introduction-to-deepagents

🗽 NYC · Jul 24 AWS AI Builder Lab: Clash of Agents Competition ‍RSVP →：https://www.aicamp.ai/event/eventdetails/W2026072406

🔁 Boston · Jul 27 LangChain Meetup: Loop Engineering & Open Models ‍RSVP →：https://luma.com/iaie4h2x

🧠 San Francisco · Jul 28 LangChain Meetup: LLM Wikis with Cognition ‍RSVP →：https://luma.com/mylwoabu

🏎️ Las Vegas · Aug 4 High Stakes, High Speed with Elastic & Google Cloud ‍RSVP →：https://events.elastic.co/f1arcadelasvegas/

🤖 Atlanta · Silicon Valley · Boston · Multiple Dates LangSmith Roadshow: Building Agents with Agents Join us for a half-day of technical sessions, live demos, an engineering leader spotlight, and a hands-on lab focused on building and improving agents in production. ‍Find your city →：https://events.langchain.com/LangSmithRoadshow/

Schneider Electric built enterprise-grade LLMOps foundations with LangSmith across 60+ AI products, giving 200+ engineers and SMEs shared observability, evaluation, and deployment standards. Read the story →：https://www.langchain.com/blog/how-schneider-electric-built-their-llmops-foundations-at-enterprise-scale-with-langsmith

Pendo used LangSmith to ship Novus, a product agent that turns user behavior and session replays into working code fixes, hitting a 90%+ success rate on PM-reviewed evals . Read the story →：https://www.langchain.com/blog/how-pendo-used-langsmith-to-trace-novus-from-user-behavior-to-code-fixes

For any additional questions or feedback, email us at hello@langchain.dev.

LangSmith, our agent engineering platform, helps developers debug every agent decision, eval changes, and deploy in one click.