All Products
Search

This Product

    Microservices Engine:What is AI Registry?

    Document Center

    Microservices Engine:What is AI Registry?

    Last Updated:Aug 24, 2026

    AI Registry (AI Governance Center) is the AI asset management platform of Alibaba Cloud Microservices Engine (MSE). It gives Agent developers registration, version management, security review, and distribution for AI assets such as Skills and Prompts.

    In short, AI Registry is your AI asset repository and distribution hub. Register the Prompts you write and the Skills you debug, and assign a version number to each. Every Agent then pulls resources from the repository by version, instead of copying files and editing configurations on each machine.

    Concepts

    AI Registry builds the following hierarchy around AI resources:

    Workspace (Namespace)
├── Prompt       → System/user prompts, few-shot examples
└── Skill        → Reusable capability unit (SKILL.md + resource files)
    Note

    The current version supports single-level organization only. Prompts and Skills attach directly to a workspace, and no further subcategory or group is supported.

    Concept

    Description

    Workspace (Namespace)

    A resource isolation container. Each workspace holds an independent set of AI assets.

    Skill

    A reusable AI capability unit that consists of a SKILL.md description file and resource files. It supports version management and public or private visibility.

    Prompt

    System prompts, user prompts, and few-shot examples. Supports versioning, tag-based retrieval, online debugging, and AI-assisted optimization.

    Each workspace contains multiple Prompts and Skills, managed and versioned independently.

    Benefits

    Security and compliance

    AI Registry reviews each Skill for security before publication, which keeps your assets compliant.

    Access control

    AI Registry integrates deeply with the Alibaba Cloud RAM (Resource Access Management) permission system and supports fine-grained access control at workspace level. A runtime permission system (Agent Identity) is planned. It assigns an independent identity to each Agent for end-to-end permission tracking and control.

    Zero maintenance

    AI Registry is a fully managed SaaS service, so you deploy and maintain no infrastructure. You register, manage, and distribute AI assets from the console without any configuration.

    Capabilities

    Skill full lifecycle management

    Stage

    Capabilities

    Create or import

    Create a Skill manually in the console, or import an existing Skill through the CLI or a ZIP upload.

    Version management

    Draft → Submit for review → Publish a version. Supports version rollback and diff comparison.

    Security review

    Automatic content safety scanning before publication, including prompt injection detection and compliance checks.

    Distribution and download

    Published versions serve consumers, which pull a Skill by version number or by the latest tag.

    Sharing and reuse

    Public or private visibility. All users can discover and use public Skills.

    Prompt debugging and optimization

    • Create a Prompt, save it as multiple versions, and retrieve versions by tag.

    • Use the built-in online debugger to send a Prompt and view the model output in real time.

    • AI-assisted optimization: AI Registry analyzes Prompt quality automatically and returns improvement suggestions.

    Multi-environment and multi-team resource isolation

    • Use workspaces to isolate resources by environment, such as development, testing, and production, or by team.

    • Each workspace manages its own Prompts and Skills independently.

    • Combine workspaces with RAM access policies for fine-grained access control.

    Scenarios

    • Shared Skills across multiple Agents — Register common Skills, such as file processing or code generation, in AI Registry. All Agents pull them on demand, which avoids duplicate development and maintenance.

    • Team Prompt version management — Prompt engineers manage multiple Prompt versions in the Registry, and Agents pull by version. Validate a new version in canary before you switch all traffic to it.

    • Cross-platform Agent integration — AI Registry acts as a unified AI asset center. Agents built on frameworks such as AgentScope and LangChain pull resources from it.

    • Enterprise-grade security and compliance — AI Registry scans each Skill for security before publication. Together with RAM permissions and audit logs, this meets enterprise security and compliance requirements.

    Access methods

    You can open the AI Registry console directly in a browser. When a client SDK, the Nacos CLI, AgentScope, or an API call accesses resources, choose one of the following methods based on your deployment environment:

    • Public network access: Suitable for local development, an IDC, or applications that do not run in an Alibaba Cloud VPC. After you enable it, the system assigns a dedicated public endpoint domain to the workspace and uses an IP address whitelist to control which public egress addresses can connect. For the configuration steps, see Configure Public Network Access for AI Registry.

    • VPC private network access: Suitable for applications that run in an Alibaba Cloud VPC. PrivateLink connects the VPC of your application to AI Registry, so access traffic does not pass through the public network. For the configuration steps, see AI Registry VPC Privatelink Configurations.

    Get started

    For the full flow, from creating a workspace to using your first Skill, see AI Registry Skill Quick Start.

    More references