DocsICP FilingConsole
Documentation
Home Page Compute Nest Use Cases Agent deployment Agent marketplace deployment

Agent marketplace deployment

Updated at:
Copy as MD
Product
My Favorites

This document describes how to select an agent from the Agent Marketplace and deploy it in one click by using Agent Deployment in Compute Nest.

What is the Agent Marketplace?

The Agent Marketplace provides official sample agents. After you select an agent and complete the required fields on the deployment page, the system automatically prepares the environment and deploys the agent. You can then interact with the agent through a Web conversation interface or by calling its HTTP API.

The same agent typically provides two deployment templates: ECS version and Container Cluster version:

  • ECS version: Runs Docker containers on a single cloud server. Suitable for quick validation and lightweight production, and supports Workbench remote logon.

  • Container Cluster version: Deploys images to an ACS/ACK cluster, where Kubernetes automatically manages the Pod lifecycle. Suitable for production environments that require high availability and elastic scaling.

Procedure

  1. Log on to Alibaba Cloud Compute Nest and navigate to Agent Deployment > Agent > Agent Marketplace.

  2. Select the agent you need and click Deploy Now.

  3. On the deployment page, select the deployment type: ECS version or Container Cluster version.

    Note

    For the Container Cluster version, you must first create a Redis instance in the target VPC and create a session connection in Agent Deployment > Connection Management. For details, see the Application connections section in this document.

  4. Complete the required fields as prompted on the page and click Create Now.

  5. After creation, you are redirected to the application page. Wait about 2 to 3 minutes (cluster creation may take longer) for the deployment to complete.

  6. On the Overview tab, open the WebUI (or use the provided access URL) to start a conversation with the agent.

Agent deployment page

Application information

  • Application name and group name: You can customize these values. In most cases, the defaults are sufficient.

Deployment resource configuration

ECS version

Parameter

Description

Billing method

Pay-as-you-go or Subscription

Instance type

Default: ecs.u1-c1m2.large

System disk

ESSD, default 40 GB

Internet bandwidth

Default 5 to 10 Mbps

Container Cluster version

Parameter

Description

Cluster options

Create ACS (default), Create ACK, or Use existing cluster

ACS

Serverless; no node management required (default option)

ACK (requires additional configuration)

Worker instance type, number of nodes, network plugin

Both deployment types access the WebUI and API through port 8000. The Web UI and the /run_sse streaming API share the same port.

Model configuration

  • Model selection: Select a model from the drop-down list. In most cases, the default selection is sufficient.

  • Model Studio API Key: Obtain an API key from Model Studio and enter it here. This key is required to call the model service.

Advanced settings

  • Environment variables: Add key-value pairs for any extra environment variables your application requires, such as business configurations. Leave blank if not needed. Collapsed by default.

Network configuration

  • ECS version: Select the zone, VPC (new or existing), vSwitch, CIDR block, and other settings as prompted.

  • Container Cluster version: You must select an existing VPC and vSwitch. Creating a new VPC is not supported.

Your VPC selection affects which application connections are available in the Tool configuration section. For example, session connections require the agent and the Redis instance to be in the same VPC network.

Tool configuration

Bind the agent to cloud resources or extended capabilities under your account.

Parameter

Description

Application connections

Bind the external capabilities that the agent depends on. The list shows only connection types compatible with the current agent (such as session management, sandbox, and knowledge base).

Model Studio MCP

Marketplace agents are automatically associated with the required MCPs, and this information is read-only. Confirm that the corresponding capabilities are enabled in Model Studio. If prompted that a capability is not enabled, follow the instructions to activate it.

Skills

If the agent supports Skills, you can enable Enable Skills and select the skills to install. This option is only displayed when the current agent supports Skills.

Application connections

Session connections are used for multi-turn conversation context persistence, backed by a Redis instance. Before you bind a session connection, create a Redis instance in the Redis console and create a connection in Agent Deployment > Connection Management. You can find detailed setup steps in the Connection Usage Guide.

  • ECS version: Optional. When creating a new VPC, session connections are not displayed; you can use the built-in local Redis instead. When selecting an existing VPC, the list shows only connections in the same VPC network, which you can bind as needed.

  • Container Cluster version: Required. You must select an existing VPC and bind a session management connection in the same VPC network.

Note

Session connections must be in the same VPC network as the agent deployment. When creating a new VPC, the VPC does not exist yet, so its network cannot be matched with Redis—session connections are therefore not displayed. Creating a new VPC is only applicable in ECS scenarios.

After creation

After deployment is complete, the following outputs are available:

Output

ECS version

Container Cluster version

WebUI access URL

http://<IP>:8000, select the corresponding application name (such as conversation)

Same as above; the address is the LoadBalancer public IP

API call example

Page provides a curl example; you can also send SSE requests to /run_sse

Same as above

Code debugging

http://<IP>:8080 (Code Server)

Same as above

Remote connection

Log on to ECS through Workbench

Not supported

Initial password

Used to log on to the ECS instance. We recommend changing the password after your first login.

Not applicable

FAQ

General

Symptom

Troubleshooting

You cannot start a conversation, or you see an authentication or key-related error.

Verify that you have entered the Model Studio API Key and that the Model Studio service is enabled.

An MCP-related error occurs.

Verify that Model Studio MCP is enabled.

The session is not persistent.

Verify that you have bound a session management connection in Tool configuration > Application connections. If you bind a connection to an existing agent, go to Application Group > Publish Management and publish again for the change to take effect.

The application connection list is empty, or the session management connection is not displayed.

Verify that you have selected an existing VPC and that the Redis instance is in the same VPC network. When creating a new VPC, session connections are not displayed. (ECS version: you can skip this and use the built-in local Redis; Container Cluster version: change your selection to an existing VPC.)

The WebUI fails to open.

Check whether the deployment is complete. The initial startup can be slow; wait a few moments and try again. Verify that the access port is 8000.

Container Cluster version

Symptom

Troubleshooting

You cannot submit the deployment, or you see a message that a connection is missing.

Verify that you have bound a session management connection in Application connections that is in the same VPC as your selected cluster VPC.

A Pod is in Pending status.

Insufficient cluster resources or scheduling restrictions.

A Pod is in CrashLoopBackOff status.

Check the Pod logs and troubleshoot image startup errors or unreachable Redis address issues.
Previous：Agent deploymentNext：Custom agent deployment