AI News
官方更新 Google Developers Blog 正文 ok 891 words

发布 Agentic Resource Discovery（ARD）规范

英文标题：Announcing the Agentic Resource Discovery specification

Agentic Resource Discovery（ARD）规范旨在为跨组织、跨平台的 AI 工具、技能和代理提供统一的发现与验证标准。它利用组织自有域名下发布的目录（catalog）和联邦注册表（registry），使智能体能够通过自然语言查询或直接获取能力，并通过加密信任元数据验证发布者身份。Google Cloud 的 Agent Registry 将原生支持 ARD，为企业提供治理、身份验证与合规保障。该规范将降低智能体生态的碎片化，帮助开发者构建更开放、安全的自主代理网络。
打开原文相关 3 · 机会 3
media11.7 KB media360.0 KB media165.3 KB media348.3 KB media64.8 KB media12.8 KB media496.8 KB media103.8 KB

中文

智能体正日益成为更庞大生态系统中的参与者。它们越来越依赖分布在团队、组织和平台上的工具、技能及其他智能体。

为了让这个生态系统规模化，智能体需要能可靠地回答三个问题：正确的功能在哪里？实际上应该使用哪个功能？以及如何验证与其连接是安全的？

如今，尚无跨组织回答这些问题的标准方式。

因此，我们宣布推出Agentic Resource Discovery (ARD)，这是一项用于在网络上发布、发现和验证 AI 能力的开放规范。与行业伙伴共同开发，ARD 使得工具和服务能够安全地共享和连接，无论其底层框架、协议或提供商如何。

与智能体生态系统的合作伙伴共同打造

智能体网络缺失的一层

设想一个负责调查实时生产事件的运营智能体。要解决问题，它可能需要查询可观测性系统、搜索工程文档、审查部署历史、开支持工单，甚至咨询专业的故障排除智能体。

虽然许多平台已经提供了自定义注册表来管理这些能力，但它们仍然碎片化并困在各自的生态系统中。这种互操作性的缺失使得智能体难以在不同工具间轻松沟通。所缺少的正是一种让智能体跨组织边界发现能力并建立信任的标准方式。

ARD 提供了这一层。它标准化了如何在组织自有域名下发布能力以及如何在联邦注册表之间索引，使任何智能体都能动态找到合适的资源。之后，ARD 便退居幕后——传递可验证的信任元数据，使智能体能够使用工具的原生协议建立直接、安全的连接。

ARD 的工作方式

这种架构依赖于两个基本元素：目录（catalogs） 和 注册表（registries）。

  • 目录（Catalogs）： 为了使资源可被发现，组织发布一个描述其可用能力的目录。因为这些目录直接托管在组织自有域名下，域名的所有权就作为身份和信任的加密基础。
  • 注册表（Registries）： 注册表充当智能体网络的搜索引擎。它们抓取已发布的目录，索引其内容并使之可搜索。当智能体提交发现请求时，注册表会返回匹配的能力以及验证发布者并在连接前建立信任所需的元数据。

下面的视频演示了 ARD 客户端完全在运行时发现并执行新能力。请在终端窗口中关注以下四个关键阶段：

  1. 发布目录： 提供者在其域名的知名路径下托管一个 ai-catalog.json 文件。该目录描述了提供者的可用能力，其中可以包含诸如 MCP 服务器、A2A 智能体、OpenAPI 工具，甚至其他嵌套目录。
  2. 发现与解析： 当客户端需要某项能力时，它可以向 ARD 注册表发起自然语言意图查询（注册表可以主动抓取并索引这些目录），或者完全绕过搜索，直接从已知合作伙伴的域名获取目录。
  3. 加密验证： 对于生产环境，发现层允许发布者附加可验证的信任元数据。无论是通过搜索还是直接获取，这都使客户端智能体或注册表能够在连接端点之前主动确认发布者的真实加密身份。
  4. 直接运行时连接： 客户端智能体动态加载所选能力，使用其原生协议或 API 与之交互，并将结果返回给用户。

将 ARD 引入 Gemini Enterprise Agent Platform

正如开放网络普及了信息访问，ARD 也将普及 AI 资源发现。Google Cloud 通过 Agent Registry 在 Gemini Enterprise Agent Platform 中提供支持——这一企业级产品实现了该愿景并成为全球联邦网络的一部分。

Agent Registry 确保企业能够可信、治理并规模化地兑现这一承诺。它为搜索、发现和托管智能体资源（包括智能体、技能、MCP 服务器和其他工具）提供了完全托管支持。此外，它允许用户将这些能力直接载入 Agent Registry，并很快将支持经过身份验证的发布者入驻。

重要的是，Agent Registry 在企业治理中发挥着核心作用。这包括分配全球唯一的命名空间 URN，执行智能体出口策略，以及固定工具和规范。它还使用 Agent Identity 验证 信任清单——ARD 的加密层，用以证明智能体真实性并满足 HIPAA 等企业合规标准。

未来几个月，Agent Platform 将原生支持 ARD，使组织能够安全地将其内部注册表连接到更广泛的网络。

开始使用 ARD

ARD 规范现已可用。参与方式：

  • 发布您的第一个目录： 按照 快速入门指南 在您的域名上托管 ai-catalog.json 文件，几分钟内让您的服务可被发现。
  • 阅读规范： 查看 规范，包括完整的模式、联邦模型、信任架构和参考实现。
  • 加入社区： 通过访问我们的 GitHub 仓库 贡献实现、提出模式更新并参与 ARD 的演进。

智能体生态系统在去中心化和开放时运行得最好。ARD 是我们为保持这一点所做的贡献。我们期待您的反馈——以及您的目录！

致谢

Agentic Resource Discovery 规范依据 Apache 2.0 许可，并基于基础性的 AI Catalog 数据模型构建。我们感谢 Linux 基金会下的 AI Catalog 工作组，以及帮助塑造该规范和参考实现的启动合作伙伴。

English / 原文

Agents are becoming participants in a much larger ecosystem. They increasingly rely on tools, skills, and other agents distributed across teams, organizations, and platforms.

For this ecosystem to scale, agents need reliable answers to three questions: Where does the right capability live? Which capability should I actually use? And how do I verify it’s safe to connect to?

Today, there is no standard way to answer those questions across organizations.

That’s why we’re announcing Agentic Resource Discovery (ARD), an open specification for publishing, discovering, and verifying AI capabilities across the web. Developed with partners across the industry, ARD allows tools and services to be securely shared and connected, regardless of their underlying framework, protocol, or provider.

Built with partners across the agent ecosystem

The missing layer of the agentic web

Consider an operations agent investigating a live production incident. To resolve the issue, it might need to query observability systems, search engineering documentation, review deployment history, open support tickets, and maybe even consult specialized troubleshooting agents.

While many platforms already feature custom registries to manage these capabilities, they remain fragmented and siloed within specific ecosystems. This lack of interoperability prevents agents from easily communicating across different tools. What’s missing is a standard way for agents to discover capabilities across organizational boundaries and establish trust in what they find.

ARD provides that layer. It standardizes how capabilities are published under an organization's own domain name and indexed across federated registries, enabling any agent to dynamically find the right resources for the job. From there, ARD steps out of the way – handing off the verifiable trust metadata so the agent can establish a direct, secure connection using the tool's native protocol.

How ARD works

This architecture relies on two primitives: catalogs and registries.

  • Catalogs: To make resources discoverable, an organization publishes a catalog describing its available capabilities. Because these catalogs are hosted directly under the organization’s own domain, ownership of that domain serves as the cryptographic foundation for identity and trust.
  • Registries: Registries act as search engines for the agentic web. They crawl published catalogs, index their contents, and make them searchable. When an agent submits a discovery request, a registry returns matching capabilities along with the metadata required to verify the publisher and establish trust before connecting.

The video below demonstrates an ARD client discovering and executing new capabilities entirely at runtime. Watch for these four key phases in the terminal window:

  1. Publishing the catalog: The provider hosts an ai-catalog.json file at a well-known path on its domain. The catalog describes the provider's available capabilities, which can include things like MCP servers, A2A agents, OpenAPI tools, or even other nested catalogs.
  2. Discovery and resolution : When a client agent needs a capability, it can either query an ARD registry using a plain-language intent (which can actively crawl and index these catalogs), or it can completely bypass search and directly fetch a catalog from a known partner's domain.
  3. Cryptographic verification: For production environments, the discovery layer allows publishers to attach verifiable trust metadata. Whether found via search or direct fetch, this enables the client agent or registry to actively confirm the publisher's true cryptographic identity before connecting to the endpoint.
  4. Direct runtime connection: The client agent dynamically loads the selected capability, interacts with it using its native protocol or API, and returns the result to the user.

Bringing ARD to Gemini Enterprise Agent Platform

Just as the open web democratized information, ARD democratizes AI resource discovery. Google Cloud supports this with Agent Registry in Gemini Enterprise Agent Platform – the enterprise-grade product that delivers on this vision and forms part of this global federated network.

Agent Registry ensures enterprises can trust, govern, and operationalize that promise at scale. It provides fully hosted support for searching, discovering, and hosting agentic resources, including agents, skills, MCP servers and other tools. Additionally, it allows users to onboard these capabilities directly onto Agent Registry, and it will soon support authenticated publisher onboarding.

Importantly, Agent Registry plays a central role in enterprise governance. This includes assigning globally unique namespaced URNs, enforcing agentic egress policies, and pinning tools and specifications. It also handles secure resource management using Agent Identity to verify the trust manifest – ARD’s cryptographic layer for proving agent authenticity and meeting enterprise compliance standards like HIPAA.

Native support for ARD will be available in Agent Platform in the coming months, allowing organizations to securely connect their internal registries to the broader network.

Get started with ARD

The ARD specification is available now. To get involved:

  • Publish your first catalog: Follow the quickstart guide to host an ai-catalog.json file on your domain and make your services discoverable in minutes.
  • Read the specification: Check out the specification, including full schemas, federation model, trust architecture, and reference implementations.
  • Join the community: Contribute implementations, propose schema updates, and participate in the evolution of ARD by visiting our GitHub repository.

The agent ecosystem works best when it is decentralized and open. ARD is our contribution to keeping it that way. We’d love your feedback – and your catalog!

Acknowledgements

The Agentic Resource Discovery specification is licensed under Apache 2.0 and is built upon the foundational AI Catalog data model. We are grateful to the AI Catalog Working Group under the Linux Foundation, as well as the launch partners who helped shape the specification and reference implementations.