# Agent 业务承载 · 日跟踪 · 2026-09-29 > 日期:2026-09-29(Asia/Shanghai)。专题正文:`intel/topics/agent-hosting.md`。原料:`/workspace/intel/archive/2026-09-29/`;对照 2026-09-28 日跟踪。 ## 摘要 - **NEW · RH OpenShell secure agent onboarding + securing AI agents:** OpenShell 定位为**开源**安全 Agent 运行时,并称属 NVIDIA **Open Agent Safety Platform**;Landlock/seccomp/netns/L7;凭证在网络边界注入;在 Podman 与 **OpenShift** 上验证;计划纳入 **Red Hat AI** 原生能力——**勿写成 OpenShell GA**(站立卷宗仍标 OpenShift AI **DP** 边界)。并行「securing the systems around them」纵深架构文。 - **industry 轻交叉 ·** OpenAI Custom GPT→**plugins/skills/MCP**(VR)——开发者平台迁移,可作 MCP/可复用能力旁注,**非企业 Agent hosting 运行时首发**。 - **续档 ·** Blueprint Alliance;Open Secure AI→LF;工业 AIoT trust;Zentrix / Edwin×AAP / AutoRAG TP。 - **不抬升:** 将 OpenShell 写成已 GA 的 MCP 网关产品,或把 OpenAI 插件退役抬成 HCI Agent 平台日。 ## 原文摘录 ### RH · OpenShell onboarding(NEW) - 归档:`redhat/why-red-hat-is-building-secure-agent-onboarding-e125232b.*` - URL:https://www.redhat.com/en/blog/why-red-hat-is-building-secure-agent-onboarding > … **OpenShell** — an **open source** project and a secure agent runtime… NVIDIA **Open Agent Safety Platform**… > … sandboxes … policy engine … gateway… **Landlock**, **seccomp**… > … validated … **Podman** and **Red Hat OpenShift**… > … integrate … into **Red Hat AI** as a native platform capability… ### RH · Securing AI agents(NEW) - 归档:`redhat/securing-ai-agents-requires-securing-the-systems-around-them-79bd2c93.*` - URL:https://www.redhat.com/en/blog/securing-ai-agents-requires-securing-systems-around-them > … Identity and authorization… Tool controls… Observability and **AgentOps**… > … **OpenShell**… **BlueField**… enforcement … independently of the agent workload… ### industry 轻 · OpenAI plugins/MCP(旁注) - 归档:`industry/openai-to-retire-custom-gpts-replace-them-with-plugins-7a013a50.*` > … **skills** … **MCP** server… ChatGPT and **Codex**…(**非 HCI Agent hosting**) ### 续档指针 - Blueprint / Open Secure AI:见 `tracking/topics/agent-hosting/2026-09-28.md`。 ### 检索缺口 - OpenShell 无独立 GA/版本矩阵通稿;`docs.redhat.com` 常 blocked。