<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
  <title>MATRIX · OpenShell + Sentry</title>
  <link>https://derekwang2002.github.io/ai-intelligence/projects/nvidia-openshell-sentry/</link>
  <description>AI 模型、Agent、工程与基础设施的重要变化，附一手来源与建议等级。</description>
  <language>zh-CN</language>
  <lastBuildDate>Mon, 02 Nov 2026 16:56:00 GMT</lastBuildDate>
  <atom:link href="https://derekwang2002.github.io/ai-intelligence/feeds/projects/nvidia-openshell-sentry.xml" rel="self" type="application/rss+xml"/>
  <item>
    <title>NVIDIA 发布 OpenShell 架构并预告 Sentry</title>
    <link>https://derekwang2002.github.io/ai-intelligence/events/ev-20261021-04/</link>
    <guid isPermaLink="false">ev-20261021-04-published</guid>
    <pubDate>Wed, 21 Oct 2026 15:58:23 GMT</pubDate>
    <category>新事件</category>
    <category>TRIAL</category>
    <description>&lt;p&gt;NVIDIA 发布以 Apache-2.0 OpenShell agent runtime 为核心的参考架构。OpenShell 将 gateway、supervisor、sandbox 与 policy prover 控制放在 agent 工作负载之外。NVIDIA 还预告 Sentry：它在 BlueField-4 的硬件隔离域执行流量检查、内存观测与隔离；Sentry 计划 2027 年初开放，OpenShell 代码目前已经可用。&lt;/p&gt;&lt;p&gt;&lt;strong&gt;为什么读：&lt;/strong&gt;OpenShell 已能试验外置策略边界，Sentry 则明确保留了尚未开放的硬件隔离限制。&lt;/p&gt;&lt;p&gt;建议等级：TRIAL&lt;/p&gt;&lt;p&gt;来源：&lt;a href=&quot;https://developer.nvidia.com/blog/how-nvidia-openshell-reference-architecture-secures-and-governs-ai-agent-workloads-at-scale/&quot;&gt;NVIDIA OpenShell reference architecture&lt;/a&gt; · &lt;a href=&quot;https://developer.nvidia.com/blog/inside-nvidia-sentry-secure-ai-agents-at-runtime/&quot;&gt;Inside NVIDIA Sentry&lt;/a&gt; · &lt;a href=&quot;https://github.com/NVIDIA/OpenShell&quot;&gt;NVIDIA OpenShell repository&lt;/a&gt;&lt;/p&gt;</description>
  </item>
</channel>
</rss>
