AgentCore is an agent building and governance platform provided by Alibaba Cloud. With three capabilities — building and running, collaboration and governance, and observability and evaluation — it covers the full lifecycle of Agents. Through multi-mode building entries, unified identity authentication, and Team collaboration orchestration, it unifies Agents from multiple sources, including self-developed Agents, open-source harnesses, and commercial SaaS Agents, on a single platform, so that Agents can be built quickly, governed reliably, and collaborated effectively. This topic describes the product positioning, core features, scenarios, and supported regions of AgentCore.
What is AgentCore
AgentCore is an agent building and governance platform that supports creating and managing Agents in a unified console. AgentCore allows connecting multiple AI model providers, integrating external tools (MCP services), and monitoring the running status of Agents in real time. AgentCore supports multi-member, multi-role team collaboration and permission management, meeting enterprise-grade security and compliance requirements.
Core features
Feature | Description |
Agent management | Creates and manages Agents. Multiple running modes are supported to meet the agent configuration requirements of different business scenarios. |
Agent templates | Creates reusable Agent configuration templates to quickly deploy the same type of Agents in batches, reducing repeated configuration costs. |
Model connection | Unifies the management of model configurations and supports flexible model switching across different Agents. |
Skill management | Creates, publishes, and manages Skills. Third-party Skills can be imported from the Skill Marketplace to extend reusable professional capabilities for Agents. |
Tool management | Integrates and manages external tools based on the Model Context Protocol (MCP), providing extended tool-calling capabilities for AI Agents. |
Credential management | Centrally manages sensitive information such as AccessKey pairs used to access external systems, preventing keys from being scattered in plaintext across configurations. |
Team collaboration | Creates teams and manages member permissions. Multi-role collaboration is supported to meet enterprise-grade permission control requirements. |
Account integration | Configures account integration with platforms such as DingTalk, WeChat, and Lark, connecting Agents to daily enterprise communication channels for conversational invocation. |
Scenarios
Digital employee: A role-based agent for internal enterprise affairs. It is created and launched through configuration on a harness, directly connected to office scenarios through IM channels, and accumulates business context through memory stores and knowledge bases, taking on tasks such as onboarding processing, document review, and ticket duty.
Intelligent customer service: A single agent service handles external consultations and business processing. The sandbox runtime scales elastically on demand to cope with traffic peaks, the identity and permission system constrains the scope of data access, and observability and evaluation capabilities support end-to-end tracing and performance tuning.
Conversational data analytics: Business questions are raised in natural language, an Agent Team decomposes the tasks, and multiple Agents connect to data sources through MCP tools to complete data retrieval, modeling, and attribution respectively, outputting analytical conclusions that directly support decision-making.
AI coding assistant: Registers an enterprise's existing heterogeneous coding Agents as-is, forms teams on demand, and completes R&D tasks through unified collaboration. R&D tasks are observable end-to-end and support intervention. Token consumption is attributable, controllable, and auditable.
Supported regions
AgentCore currently supports the following regions:
Region name | Region ID |
China (Hangzhou) | cn-hangzhou |
China (Beijing) | cn-beijing |
China (Shanghai) | cn-shanghai |
China (Shenzhen) | cn-shenzhen |
Singapore | ap-southeast-1 |
Fees for dependent products
When AgentCore is used for the first time, the service activation process must be completed. During service activation, the system requires activating the following cloud products in sequence. All dependent products are billed independently according to their own billing rules and are not included in the free scope of AgentCore.
Product | Description | Activation method |
CMS (CloudMonitor) | Writes tracing data. | Go to the CloudMonitor console to activate. |
PrivateLink | Provides private network access capabilities. | Activated automatically upon click. |
SchedulerX3 | Provides distributed task scheduling capabilities. | Go to the buy page to purchase. |