For production agents, model choice is only one part of improving agent performance. Teams also need to control the system around the model, including the tools an agent can use, the context it sees, how it is evaluated, where it runs, and what policies apply to each action.对于生产环境中的智能体而言，模型选择只是提升性能的环节之一。团队还需要管控模型周边的系统，包括智能体可使用的工具、其所见的上下文、评估方式、运行位置以及适用于每项操作的策略。
Today, we’re announcing the NemoClaw for LangChain Deep Agents blueprint, developed with NVIDIA to help enterprises build open, governed agent systems. The blueprint brings together LangChain Deep Agents Code, NVIDIA Nemotron 3 Ultra, and NVIDIA OpenShell runtime so teams can tune agents for their own workloads, run them securely, and optimize for quality, cost, and speed.今天，我们正式宣布推出与 NVIDIA 共同开发的 NemoClaw for LangChain Deep Agents 蓝图，旨在帮助企业构建开放且受控的智能体系统。该蓝图整合了 LangChain Deep Agents 代码、NVIDIA Nemotron 3 Ultra 以及 NVIDIA OpenShell 运行时，使团队能够针对自身工作负载调整智能体、确保运行安全，并实现质量、成本与速度的最优化。
In our evals, Nemotron 3 Ultra with a tuned LangChain Deep Agents harness provides advanced agent performance at a much lower inference cost. The main takeaway is that agent performance improves when the model, harness, evals, and runtime are tuned together.在我们的评估中，使用经过调优的 LangChain Deep Agents 架构的 Nemotron 3 Ultra，以极低的推理成本提供了卓越的智能体性能。核心结论是：当模型、架构、评估体系和运行时协同调优时，智能体性能将得到显著提升。
An open agent stack for enterprise workloads面向企业级工作负载的开放式智能体栈
As enterprises move agents into production, the systems they build around the model become valuable IP. Agent memory, workflows, traces, eval datasets, harness configuration, and tuning data all reflect a company’s unique domain expertise. This is proprietary knowledge that can shape how the company competes, but in closed ecosystems, teams don’t fully control it. They need a way to own that work, improve it over time, and run agents with the controls their organizations require.随着企业将智能体推向生产环境，围绕模型构建的系统已成为极具价值的知识产权。智能体的记忆、工作流、追踪记录、评估数据集、架构配置以及调优数据，都体现了企业独特的领域专业知识。这是能够塑造企业竞争力的专有知识，但在封闭的生态系统中，团队无法完全掌控。他们需要一种方式来拥有这些成果，实现持续改进，并以组织要求的管控标准运行智能体。
The NemoClaw for Deep Agents blueprint gives teams control over the full agent stack.NemoClaw for Deep Agents 蓝图赋予团队对整个智能体栈的完全控制权。
- An open model layer. Nemotron 3 Ultra gives teams an open model they can run, customize, and optimize for enterprise workloads.开放的模型层。Nemotron 3 Ultra 为团队提供了一个开放模型，他们可以对其进行运行、定制和优化，以适配企业级工作负载。
- A tuned agent harness. LangChain Deep Agents Code (dcode) provides the harness layer for long-running agents, including planning, tool use, memory, and task execution. The blueprint includes a Deep Agents harness profile tuned for Nemotron 3 Ultra.调优后的智能体架构。LangChain Deep Agents 代码 (dcode) 为长期运行的智能体提供了架构层，涵盖规划、工具使用、记忆和任务执行等功能。该蓝图包含了一个专为 Nemotron 3 Ultra 调优的 Deep Agents 架构配置。
- A governed runtime. NVIDIA OpenShell provides a secure runtime for sandboxed agent execution, with policies for how agents interact with tools, systems, and data.受控的运行时。NVIDIA OpenShell 为智能体执行提供了安全的沙箱运行时，并针对智能体如何与工具、系统及数据交互制定了策略。
Together, these layers provide a path to build and deploy agents teams can measure, govern, and improve in production.这些层级共同提供了一条路径，帮助团队构建和部署可在生产环境中进行衡量、治理和改进的智能体。
Reaching benchmark-leading performance at 10x lower cost以 1/10 的成本达到基准测试领先性能
In LangChain’s agent eval suite, NVIDIA Nemotron 3 Ultra evaluated with LangChain Deep Agents achieved an aggregate score of 0.86 at a cost of $4.48. The next closest performing model cost $43.48, making Nemotron 3 Ultra roughly 10x lower inference cost on this benchmark.在 LangChain 的智能体评估套件中，使用 LangChain Deep Agents 进行评估的 NVIDIA Nemotron 3 Ultra 取得了 0.86 的总分，成本仅为 4.48 美元。性能最接近的另一款模型成本为 43.48 美元，这意味着在该基准测试中，Nemotron 3 Ultra 的推理成本降低了约 10 倍。
To reach these results, we tuned how the agent uses tools, manages context, and evaluates intermediate steps. The goal was to adapt the harness around Nemotron 3 Ultra’s strengths and the common patterns that show up in long-running agent tasks.为了达到这些结果，我们对智能体使用工具、管理上下文以及评估中间步骤的方式进行了调优。目标是根据 Nemotron 3 Ultra 的优势以及长期运行的智能体任务中常见的模式，来适配其架构。
For enterprise teams, this means the entire agent system can be optimized around the requirements of their specific workloads. Teams fine tune model weights, customize the harness, run evals, and control the runtime based on quality, cost, latency, and governance requirements. This gives companies a way to own the agent’s core intelligence and decide when, how, and why the system changes over time.对于企业团队而言，这意味着整个智能体系统可以围绕其特定工作负载的需求进行优化。团队可以根据质量、成本、延迟和治理要求，对模型权重进行微调、定制架构、运行评估并控制运行时。这使企业能够掌握智能体的核心智能，并决定系统何时、如何以及为何随时间演进。
Why lower inference cost leads to better agents为什么降低推理成本能带来更好的智能体
Lower inference cost is one of the biggest benefits of using open models. It reduces serving cost for end users, and it also changes how teams build and improve agents.降低推理成本是使用开放模型最大的优势之一。它不仅降低了终端用户的服务成本，还改变了团队构建和改进智能体的方式。
Evals work best when teams run them throughout the agent development lifecycle. Before deployment, teams need to test changes to prompts, harnesses, tools, models, and data. After deployment, they need to monitor real behavior, make fixes needed, and create tests so the regression doesn't happen again.当团队在整个智能体开发生命周期中持续运行评估时，效果最佳。在部署前，团队需要测试对提示词、架构、工具、模型和数据的修改；在部署后，他们需要监控实际行为、进行必要修复，并创建测试以防止回归。
When each iteration is expensive, teams run fewer evals, compare fewer variants, and avoid specialized agents because the operating cost is too high.如果每次迭代的成本都很高，团队就会减少评估次数、对比更少的变体，并因运营成本过高而避免使用专业化智能体。
A more cost-efficient open stack makes it practical to:更具成本效益的开放栈使得以下操作变得切实可行：
- Run larger eval suites before deployment and in production在部署前及生产环境中运行更大规模的评估套件
- Compare more model, harness, and tool variants对比更多的模型、架构和工具变体
- Evaluate specialized agents for specific domains针对特定领域评估专业化智能体
The best system for a given workload optimizes across quality, cost, speed, and governance together. For example, the latency requirements for real-time customer support agents look very different from a coding agent running concurrent tasks in the background.针对特定工作负载的最佳系统，是能够同时在质量、成本、速度和治理方面实现优化的方案。例如，实时客户支持智能体的延迟要求，与后台运行并发任务的编码智能体有着显著差异。
“Super agents have arrived. With an open model like NVIDIA Nemotron, a LangChain harness, the NVIDIA OpenShell runtime, and a company’s own data, every enterprise can build custom agents that understand its business, use its tools, and turn knowledge into action. The future of AI won’t be one-size-fits-all — companies will use AI cloud services and build their own AI, shaped by their proprietary data, know-how, and workflows, and run it safely and securely wherever they operate.”
– Jensen Huang, Founder and CEO of NVIDIA“超级智能体时代已经到来。借助 NVIDIA Nemotron 等开放模型、LangChain 架构、NVIDIA OpenShell 运行时以及企业自身的数据，每家企业都能构建出理解其业务、使用其工具并将知识转化为行动的定制化智能体。AI 的未来不会是‘一刀切’的——企业将利用 AI 云服务并构建属于自己的 AI，这些 AI 由其专有数据、专业知识和工作流所塑造，并能在其运营的任何地方安全、稳妥地运行。”—— NVIDIA 创始人兼首席执行官黄仁勋 (Jensen Huang)
Ecosystem support生态系统支持
The announcement is supported by partners across the AI infrastructure and enterprise ecosystem, including EY, who is building an implementation practice around the software stack, and Baseten, Fireworks, Nebius, Crusoe, DeepInfra, and Together AI. These partners help enterprises serve Nemotron models in production and adapt the blueprint for business critical applications.该公告得到了 AI 基础设施和企业生态系统合作伙伴的支持，包括正在围绕该软件栈构建实施实践的 EY（安永），以及 Baseten、Fireworks、Nebius、Crusoe、DeepInfra 和 Together AI。这些合作伙伴帮助企业在生产环境中部署 Nemotron 模型，并针对关键业务应用适配该蓝图。
“EY clients in regulated industries are ready to move agentic AI out of isolated pilots and into production and are often constrained by governance, security, and the ability to prove control to a regulator or a board. Open agent architectures matter because they give enterprises transparency into how agents operate, control over where data and inference run, and the freedom to deploy on their own terms without committing to a closed stack. By delivering the NVIDIA NemoClaw blueprint, which incorporates LangChain technology, EY teams help give clients a secure, sandboxed foundation for always-on agents that can meet enterprise standards for auditability and risk from the first deployment."
– Geoff Vickrey, Global Chief Commercial Officer, NVIDIA, EY“安永在受监管行业的客户已准备好将智能体 AI 从孤立的试点项目推向生产环境，但往往受限于治理、安全以及向监管机构或董事会证明合规的能力。开放式智能体架构之所以重要，是因为它们为企业提供了智能体运行方式的透明度、对数据和推理运行位置的控制权，以及无需绑定到封闭栈即可按自身条款进行部署的自由。通过提供集成 LangChain 技术的 NVIDIA NemoClaw 蓝图，安永团队正帮助客户为始终在线的智能体提供安全、沙箱化的基础，从首次部署起即可满足企业在可审计性和风险方面的标准。”—— Geoff Vickrey，NVIDIA 全球首席商务官，安永 (EY)
"Production agents need inference that is fast, reliable, and cost-efficient at scale. We have optimized NVIDIA Nemotron models on Baseten to deliver high throughput and low latency on NVIDIA hardware, so teams get strong price-performance without operating the infrastructure themselves. Delivering Nemotron through the NemoClaw blueprint with LangChain gives enterprises a clear path to run open agentic models in production with the performance and economics these workloads demand."
– Philip Kiely, Head of Developer Relations, Baseten.“生产环境中的智能体需要快速、可靠且具成本效益的大规模推理。我们在 Baseten 上优化了 NVIDIA Nemotron 模型，以在 NVIDIA 硬件上提供高吞吐量和低延迟，从而让团队在无需自行运维基础设施的情况下获得强大的性价比。通过 NemoClaw 蓝图与 LangChain 提供 Nemotron，为企业提供了一条在生产环境中运行开放式智能体模型的明确路径，并满足了这些工作负载所需的性能与经济性。”—— Philip Kiely，Baseten 开发者关系主管。
“Agentic workloads make many model calls per task, so inference speed and cost directly determine whether an agent is viable in production. Fireworks serves NVIDIA Nemotron models with the throughput and price-performance that high-volume agent systems require, tuned for the tool calling and reasoning patterns these workloads depend on. Offering Nemotron through the NemoClaw blueprint with LangChain gives enterprises an efficient, open foundation they can scale with confidence."
– Lin Qiao, CEO and Cofounder, Fireworks AI.“智能体工作负载在每个任务中会进行多次模型调用，因此推理速度和成本直接决定了智能体在生产环境中是否可行。Fireworks 以高容量智能体系统所需的高吞吐量和性价比提供 NVIDIA Nemotron 模型，并针对这些工作负载所依赖的工具调用和推理模式进行了调优。通过 NemoClaw 蓝图与 LangChain 提供 Nemotron，为企业提供了一个高效、开放且可自信扩展的基础。”—— Lin Qiao，Fireworks AI 首席执行官兼联合创始人。
“The next challenge for enterprise AI is running complex agentic workloads economically at production scale. Nebius was built for that challenge. Our AI-native cloud gives customers dedicated infrastructure optimized for high-performance inference and cost-efficient scaling. By offering NVIDIA Nemotron models through the NemoClaw blueprint with LangChain, we’re making it easier for organizations to deploy and scale open agentic AI across their business.”
– Roman Chernin, Chief Business Officer, Nebius“企业 AI 面临的下一个挑战是在生产规模下经济地运行复杂的智能体工作负载。Nebius 正是为此而生。我们的 AI 原生云为客户提供了专为高性能推理和经济高效扩展而优化的专用基础设施。通过 NemoClaw 蓝图与 LangChain 提供 NVIDIA Nemotron 模型，我们正在让各组织能够更轻松地在业务中部署和扩展开放式智能体 AI。”—— Roman Chernin，Nebius 首席商务官
Availability可用性
The NemoClaw for LangChain Deep Agents blueprint is available today.NemoClaw for LangChain Deep Agents 蓝图现已发布。
For deeper technical detail, read:欲了解更多技术细节，请阅读：
- Deep Agents Code on NemoClaw: a governed blueprint for your most sensitive codeNemoClaw 上的 Deep Agents 代码：为您最敏感的代码提供的受控蓝图
- Tuning the harness, not the model: a Nemotron 3 Ultra playbook调优架构而非模型：Nemotron 3 Ultra 实战指南
