AgentBay:Service support

Q1: What is AgentBay? Is it a virtual machine?

AgentBay is more than a virtual machine. It is a comprehensive cloud infrastructure platform designed for AI Agents. Key aspects include the following:

• Positioning: It is a cloud infrastructure for AI Agents, launched by Alibaba Cloud for enterprises, developers, and AI vendors.

• Core features:

  • Standard runtime: AgentBay pre-integrates many standard tools for Agent task execution. These tools are encapsulated using the Model Context Protocol (MCP) for rapid enterprise integration.

  • User state persistence: AgentBay uses a proprietary persistence architecture to save user profiles, cookies, and other data in an isolated and secure manner. This data is dynamically mounted to make the cloud environment closely resemble an on-premises environment.

  • Real-time client-cloud interaction: AgentBay uses the proprietary Alibaba Cloud Adaptive Streaming Protocol (ASP) to stream the cloud display to the user's on-premises device in real time. This allows both the user and the AI to take turns controlling the cloud environment. ASP also supports network and peripheral redirection channels, which allows the cloud environment to sense and control on-premises devices.

• Technical architecture:

  • A proprietary persistent file system that can dynamically retain user states and files on demand.

  • A proprietary client-cloud real-time communication protocol ASP that allows the cloud environment to sense, imitate, and control the on-premises environment.

  • Built on Alibaba Cloud Workspace resource pools that are deployed globally for low-latency access.

  • Provides serverless capabilities, allowing customers to schedule environment sessions with a single click.

  • Integrates standard MCP toolsets such as Browser, File, and Terminal.

  • Supports fast integration via the SDK/API or the MCP Server. Both methods are fully supported and compatible.

Q2: How can AI Agent products on the market connect to AgentBay?

AI Agent products can connect to AgentBay in the following ways:

1. Basic connection flow:

   1. Request an API key.

   2. Create a custom image and configure the resource pool definition.

   3. (Optional) Configure the MCP service.

   4. (Optional) Integrate the Alibaba Cloud Workspace web SDK.

   5. Initiate a tool use request (session scheduling) from the client or server.

2. Connection types:

   • Connect using the Alibaba Cloud Workspace server-side SDK and API.

   • Connect using an MCP Server.

3. Access methods:

   • Supports streaming the cloud environment through an embedded WebView in the client.

   • Supports redirection to a browser for use.

   • Access URL format: https://wuying.aliyun.com?mcp.html?authcode=<authCode>&resourceId=<resourceId>

     Note

     When you initiate a tool call, the MCP Server returns this URL.

Q3: AgentBay supports the MCP protocol. What is its relationship with MCP Servers on the market?

The relationship between AgentBay and an MCP Server is as follows:

• Protocol support:

  • AgentBay fully supports the MCP.

  • AgentBay provides standardized MCP interfaces to ensure compatibility with the existing MCP ecosystem.

• Tool integration:

  AgentBay provides mainstream MCP toolsets, including but not limited to:

  • Browser tool: Supports browser operations.

  • File tool: Supports file operations.

  • Terminal tool: Supports terminal operations.

• Ecosystem engagement:

  • AgentBay can be seamlessly integrated with existing MCP Server ecosystems.

  • AgentBay supports feature extensions for existing MCP Servers.

  • AgentBay provides unified interface standards for easy management and integration.

Q4: When using an AI Agent, I do not want to leak my personal data. How does AgentBay protect my private data?

AgentBay uses the following mechanisms to protect user privacy and data security:

• Isolation mechanism:

  • AgentBay provides a completely isolated virtual machine environment for each user.

  • The environment is reset after each session ends, and no data is retained.

• Access control:

  • AgentBay uses API keys for identity authentication.

  • AgentBay provides secure access control mechanisms.

• Data processing:

  • On-premises data is not uploaded to the cloud, and cloud data is not retained, unless you explicitly grant authorization and make a request.

  • Provides a secure channel for data transmission.

  • AgentBay supports encrypted data storage.

Q5: Agents often consume a lot of on-premises computing power when executing tasks, preventing my computer from doing other things. Can AgentBay solve this problem? How?

AgentBay can solve the problem of high on-premises computing power consumption in the following ways:

• Cloud execution:

  • Tasks run in Alibaba Cloud resource pools and do not consume on-premises computing resources.

  • AgentBay provides a serverless service that allocates computing resources on demand.

  • AgentBay supports scalability to automatically adjust resources based on task requirements.

• Resource optimization:

  • AgentBay uses a cloud-based distributed computing architecture.

  • AgentBay supports parallel processing for tasks.

  • AgentBay automatically scales resources in or out.

• Performance guarantee:

  • AgentBay provides professional cloud computing resource pools.

  • Ensures efficient task execution.

  • On-premises devices only need to handle basic interactive operations.

Q6: What specific Desktop Auto capabilities does AgentBay currently offer? Is it supported to use custom images to connect to AgentBay?

AgentBay provides a rich set of Desktop Auto capabilities:

• Basic automation capabilities:

  • Browser tool: Supports complete browser automation operations.

  • File tool: Supports file system automation operations.

  • Terminal tool: Supports command-line automation operations.

  • Python tool: Supports secure code execution within the environment.

• Advanced automation capabilities:

  • GUI Agent: Provides the ability to understand real-time screen content and execute interactive instructions.

  • Custom knowledge space: Supports automatic accumulation of user knowledge and custom saving of user states.

  • Supports the development of custom application interfaces.

• Image support:

  • Provides a built-in Linux base image.

  • Supports the creation of custom images based on the base image.