NVIDIA 扩展 Agent Toolkit 工具包,强化智能体对企业数据的访问控制NVIDIA AI Says Expanded Toolkit Controls Agent Data Access


AI free token/AI免费积分
NVIDIA 扩展 Agent Toolkit 工具包,强化智能体对企业数据的访问控制

NVIDIA AI 于 2026 年 11 月 4 日宣布,扩展后的 NVIDIA Agent Toolkit 让开发者能够构建这样的智能体:在企业数据的访问控制之下,用这些数据来回答问题、做出预测。NVIDIA 在 X 平台的官方账号写道:"我们扩展了 NVIDIA Agent Toolkit,帮助开发者构建能利用企业数据回答问题、做出预测的智能体,并控制智能体可访问与可执行的范围。"该帖同时点名 Lila Sciences、Siemens(西门子)和 Dassault(达索)正在用 AI 智能体设计潜在疗法、规划工厂生产与优化车辆设计。

这套工具包的源头是 NVIDIA 博客于 10 月 21 日由 Erik Pounds 发布的文章《NVIDIA Agent Toolkit Expands, Grounding Enterprise Agents in Company Data》。文章开宗明义:"智能体有多大用,取决于它知道什么;有多可信,取决于它运行其中的边界有多严密。"文章列出的工具包包含以下新组件。

一、NVIDIA Agent Knowledge Foundation(智能体知识基础架构):帮助企业把数据转化为智能体可用的"企业智能"。智能体可找到相关信息、理解其对业务的意义,并用它来做预测、辅助决策。企业无需迁移数据、重建现有工作流或改变治理控制即可使用。它整合了 NVIDIA CUDA-X 系列库(cuDF、cuGraph、cuML、cuOpt、cuVS)以及全新的 Auto-Ontology 库——后者将原始企业数据翻译成客户、产品、订单等熟悉的业务概念,并刻画它们之间的关联。

二、NVIDIA 开放模型:包括 Nemotron Retriever 检索模型,以及全新的 Kumo 系列结构化开放模型——Kumo Tabular、Kumo Relational、Kumo Time Series 和 Kumo Forecast,用于帮助智能体找到相关企业信息、回答问题并做预测。

三、NVIDIA OpenShell:被描述为一个"安全的智能体运行时(secure agent runtime)",为智能体提供受控的运行边界,限制其可访问与可执行的范围。

四、NVIDIA NemoClaw:提供构建安全、专用智能体的参考蓝图(blueprint);其中最新的"智能体优化蓝图"给出一步步的起点,帮助智能体随着时间推移把本职工作做得更好。

五、工具与技能:如 NeMo Relay 和 NeMo Compass,为部署与优化提供起点。

Kumo 各模型分工明确:Kumo Tabular 面向结构化表格的预测;Kumo Relational 跨关联表与实体做预测;Kumo Time Series 分析按时间排序的观测;Kumo Forecast 预测未来数值与业务指标。Kumo Relational 可跨已关联的表运行,从而降低把记录拍平到单一训练表或特征库所需的预处理工作量。

在 10 月 21 日的文章中,NVIDIA 以自身口吻描述了 Agent Knowledge Foundation 的三处落地:与 Palantir 合作,NVIDIA 供应链部门正通过 Palantir Foundry、Ontology 与 AIP 部署专用智能体,从专有数据生成可执行情报、预测结果并优化决策;Databricks 将 NVIDIA Kumo Anomaly 与 Databricks Genie Ontology 配对,驱动具备上下文感知的智能体;Snowflake 正让客户在 Snowflake Notebook 环境中使用 NVIDIA Kumo Engine 与 Kumo Relational 开放模型,释放更强的预测能力。

搜索与预测共享同一层:Nemotron Retriever 负责生成 Embedding 并执行语义搜索的重排序(reranking)。多家合作伙伴报告了可观收益——Amazon OpenSearch Service 借助其 Embedding 以更少搜索步骤、几乎零额外延迟提升了相关性;SAP 微调后的重排序模型速度快了近 3 倍;ServiceNow 在重排序质量提升的同时,将延迟降低了近四分之一。

运行时边界方面,OpenShell 为会与系统交互、调用工具或修改记录的智能体提供安全运行时边界,可限制其访问与操作;但团队仍需自行定义具体权限、管理敏感凭据、记录活动,并测试策略在工具出错时的表现。NemoClaw 蓝图则提供专用智能体的参考架构,在保持对执行过程显式控制的同时,支撑评估与迭代优化。

从更宏观的视角看,NVIDIA 正把本体构建、GPU 加速处理、检索、结构化预测与安全执行打包成一个连贯的技术栈,意在填补现有系统(LangChain、LlamaIndex 等编排框架,以及数据仓库、向量库、云平台)与智能体应用之间的"中间空白"。对已经运行 NVIDIA GPU 与受支持企业数据平台的机构,这套方案收益最直接。

综上,扩展后的 NVIDIA Agent Toolkit 通过 Agent Knowledge Foundation、OpenShell、NemoClaw 与 Kumo 模型族,把"企业数据可用、可治理、可控"落到实处:智能体既能基于公司数据回答与预测,又被关在明确的访问边界之内。

本文来源:https://theagenttimes.com/articles/nvidia-ai-says-expanded-toolkit-controls-agent-data-access-84ff7ab2
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

NVIDIA AI said on Nov. 4, 2026, that the expanded NVIDIA Agent Toolkit lets developers build agents that answer questions and make predictions from company data under access controls. NVIDIA's account on X wrote: 'We expanded the NVIDIA Agent Toolkit to help developers build agents that use company data to answer questions and make predictions, with controls over what agents can access and do.' The post names Lila Sciences, Siemens and Dassault as using AI agents to design potential therapies, plan factory production and refine vehicle designs.

The toolkit originates from an Oct. 21 NVIDIA Blog post by Erik Pounds, 'NVIDIA Agent Toolkit Expands, Grounding Enterprise Agents in Company Data.' Its first sentence: 'Agents are only as useful as what they know, and only as trustworthy as the boundaries they run within.' The toolkit now includes several new components.

NVIDIA Agent Knowledge Foundation helps enterprises turn data into company intelligence agents can use. Agents can find relevant information, understand what it means for the business and use it to make predictions and help people decide what to do next. Companies can use it without moving data, rebuilding existing workflows or changing governance controls. It features NVIDIA CUDA-X libraries (cuDF, cuGraph, cuML, cuOpt, cuVS) and the new Auto-Ontology library, which translates raw company data into business concepts like customers, products and orders and maps how they relate.

NVIDIA open models include Nemotron Retriever and the new Kumo family of structured open models: Kumo Tabular, Kumo Relational, Kumo Time Series and Kumo Forecast, which help agents find relevant company information, answer questions and make predictions.

NVIDIA OpenShell provides a secure agent runtime, giving enterprises control over what agents can access and do. NVIDIA NemoClaw provides reference designs for building secure, specialized agents; the newest NemoClaw blueprint for agent optimization offers a step-by-step starting point for improving agent intelligence over time.

NVIDIA AI Says Expanded Toolkit Controls Agent Data Access

Tools and skills such as NeMo Relay and NeMo Compass provide supporting agent tools and starting points for deployment and optimization. Kumo Tabular predicts over structured tables; Kumo Relational predicts across connected tables and entities; Kumo Time Series analyzes time-ordered observations; Kumo Forecast forecasts future values and business metrics.

Early deployments named in the post: NVIDIA's supply chain organization uses Agent Knowledge Foundation with Palantir Foundry, Ontology and AIP to deploy specialized agents and predict outcomes; Databricks pairs NVIDIA Kumo Anomaly with Databricks Genie Ontology to power context-aware agents; Snowflake enables customers to use the NVIDIA Kumo Engine and Kumo Relational open models in Snowflake Notebook environments. Partner-reported gains include Amazon OpenSearch Service improving relevance with fewer search steps, SAP's fine-tuned reranker running nearly 3x faster, and ServiceNow cutting reranking latency by nearly 25%.

Together, the expanded toolkit grounds enterprise agents in company data while keeping them inside explicit access boundaries: agents can reason over proprietary information, but only within the controls the enterprise defines.

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry