Activate Cluster

Activate Cluster is the entry point for bringing your GPU infrastructure into Smart Studio. The guided workflow installs the platform agent, verifies every machine, and deploys the Smart Studio platform to your servers.

You provide	Smart Studio handles	Successful result
CPU and GPU servers, network access, and machine credentials	Agent connection, compatibility checks, and platform deployment	An Activated cluster with access to GPU Dashboard and model deployment

You can leave and come back

Your progress is saved automatically. You can close the page at any time and resume from the same step later. You can also run multiple activations in parallel from the Activate Service list.

Prerequisites

Complete the following prerequisites:

• If you are deploying the Model Serving Platform (BYO-GPU): Review the BYO-GPU Requirements and confirm your infrastructure meets the minimum requirements.
• If you are deploying the API Router Platform (BYO-Key): Review the API Router Platform Requirements and confirm your infrastructure meets the minimum requirements.
• Confirm your Alibaba Cloud account can authorize the OSS, VPC, and ECI products.
• Keep the CPU server connected to the internet throughout the activation.

Step 0: Choose the Billing Model

1. Review the billing rules and keep the agreement checkbox selected.
2. Choose the billing model that fits your use case.

	Model Service Platform	API Router Platform
Mode	BYO-GPU	BYO-Key
Use case	Deploy models on your own GPUs to get tokens for internal use or integrate into an existing platform	Resell Model APIs in white-labelled User Console and Admin Console
Billing model	Option A: Token Sharing Sharing rate: 15% Option B: Fixed Amount Fee According to GPU Spec. Free until August 15th.	Option A: Token Sharing Sharing rate: Model Studio 6% / Other providers 8%

warning

The billing model cannot be changed after it is confirmed. Read the billing rules carefully before you make your selection.

Continue to the Next Step

Confirm your selections, click Confirm and Activate, then proceed to Step 1.

Step 1: Prepare the Environment

Read and understand the deployment architecture and install the Smart Studio agent on your CPU server.

Review the Reference Architecture

Choose the platform you want to deploy (Model Service Platform / API Router Platform), open the Architecture Overview panel and choose the topology that fits your environment: standard deployment or POC deployment. For a detailed walkthrough, see the BYO-GPU Requirements. If you are deploying the API Router Platform, refer to the API Router Platform Requirements for CPU-only infrastructure specifications.

Install the Agent

1. Copy the installation command from the page. Your server address and activation token are unique to your account.

2. Run the command on your CPU server:

   curl -fsSL http://<server-address>/download/install.sh | bash -s -- \
     --server http://<server-address> \
     --activation-token <your-activation-token>

3. Keep the terminal session open. Wait for the agent to report running (systemctl status xt-agent).

Note

Use the exact command from your console. Do not modify the server address or token.

Continue to the Next Step

Click Completed and Next. Smart Studio verifies the agent connection and advances to Step 2.

If the agent is not detected, see The agent isn't detected.

Step 2: Provide Server Information

Enter the connection details for your servers.

Enter Machine Configuration

1. Paste your server details into the Machine Configuration text area. Use the format shown on the page.

2. To add multiple servers, add one block per server.

   Example:

    # Machine 1
    public ip: xxx
    ip: xxx
    port: xxx
    username: root
    password: your_password
   
    # Machine 2
    public ip: xx
    ip: xxx
    port: XXX
    username: root
    password: your_password

3. Confirm every entry includes the ip:, username: and password: fields. The activation process rejects incomplete entries.

Continue to the Next Step Click Submit for Analysis. Then the AI agent analyzes your configuration and the process advances to Step 3 automatically.

If you want to change the information later, return to this page to adjust your configuration.

Step 3: Verify Machine Compatibility

AI Agent will run compatibility checks on each server in parallel and it will take several minutes to do the verification.

Understand the Status Labels

The page shows one card per server. Each card includes the server IP address, an overall status label, and the result of each check.

Label	Meaning
Passed	The server meets all requirements.
Checking…	Verification is in progress.
Issues	One or more checks failed on this server.

All buttons are disabled during verification.

Wait for Verification to Complete

When every server reports Passed, the wizard advances to Step 4 automatically.

Handle Failed Checks

If any server reports Issues, choose one of the following actions:

• Click Edit Information to return to Step 2 and update the server details.
• Click Resubmit to rerun the checks without changing the configuration. Use this option for transient network issues.

If any check fails, see Machine verification reports failed checks. If verification stops with a Connection Failed status, see Agent connection is lost during verification.

Step 4: Deploy the Platform

Once the previous steps are complete, you can start to deploy the platform on your servers. Deployment typically takes 60+ minutes and varies with your server configuration (larger memory usually takes longer). The page displays real-time deployment status:

Start Deployment

In Progress

Activation Success

If deployment fails, see Platform deployment fails.

Manage Activated Clusters

After activation, you will move to the Activate Service list. From this list, you can resume any in-progress activation, Activate More Clusters, or stop a running service with Deactivate Service.

Deactivation is disruptive

• All deployed Smart Studio software services and model services become unavailable after deactivation.
• The minimum guarantee fee & fixed amount fee stops on T+1.

Access the Platform

After Smart Studio is successfully deployed, your chosen platform is running on your servers. The IP address used to access the platform matches the machine configuration you provided in Step 2: Provide Server Information. You can log in with the default username and initial password.

If you chose to deploy the API Router Platform

You can access the platform using your server's IP address:

• Admin Console: https://<Your-IP>/admin
• User Console: https://<Your-IP>/

From the Admin Console, administrators can:

• Onboard models — List models for sale on the platform.
• Set model pricing — Configure both the cost price and the selling price for each model.
• Manage customers — Manage customer accounts, discounts, and rate limits.
• View bills — Check usage statistics and billing details.

From the User Console, end-users can:

• Top up — Add balance to their account to use platform services.
• Call models — Invoke models listed on the platform via API.
• Try models — Experience model chat and generation capabilities online.

If you chose to deploy the Model Serving Platform (MSP)

You can access the MSP using the server's IP address with the /model-serving path. For example, if your server's internal IP is 192.168.0.1, users within the same VPC can access the MSP at http://192.168.0.1/model-serving.

From the MSP console, you can:

• Deploy models — Deploy open-source models to your GPU cluster.
• Manage serving services — Start, stop, scale, and monitor model services.
• Monitor performance — Track inference latency, throughput, and resource utilization.

Troubleshooting

Use the following sections to resolve common activation issues.

The Agent Isn't Detected

After you click Completed and Next, the activation token can't be validated. This usually means the agent didn't start on your CPU server, or the token expired.

To fix it:

1. Confirm the installation command finished on your CPU server.
2. If the agent isn't running, reinstall it using the exact command shown in your console.
3. If the token expired, refresh the page to get a new command, then reinstall the agent.
4. Click Completed and Next again.

Machine Verification Reports Failed Checks

One or more servers show the Issues label, it means a check did not meet the compatibility requirement. Common causes include an unreachable IP, missing GPU driver, closed port, or insufficient SSH permission.

To fix it:

1. Expand the failed server card to see which check failed.
2. For incorrect server details, click Edit Information to return to Step 2 and correct the configuration.
3. For transient network issues, click Resubmit to rerun the checks.
4. If the failure persists, review your infrastructure against the BYO-GPU Requirements.

Platform Deployment Fails

The deployment page shows a Partial failure state with a Retry Deployment button. One or more platform components failed to deploy. A temporary resource issue on the target servers is the most common cause.

To fix it:

1. Click Retry Deployment to restart the deployment.
2. If deployment fails repeatedly, verify your servers still meet the BYO-GPU Requirements requirements.
3. Contact support if the failure persists.

Agent Connection Is Lost During Verification

If Step 3 shows a Connection Failed status and verification stops, meaning the agent installed in Step 1 is no longer reachable.

To fix it:

1. On your CPU server, confirm the agent process is still running.
2. If the agent stopped, reinstall it using the exact command from your console.
3. Return to Step 3 and click Resubmit to rerun the checks.

Video Tutorial

Watch the following video for a step-by-step walkthrough of the self-service activation process.

Verify the Result

Activation is complete when:

• The activation record shows Activated.
• Connection Status shows Connected.
• The GPU Dashboard action is available.
• The expected nodes and GPU capacity appear in the dashboard.

Open GPU Dashboard before creating a model deployment. This confirms that Smart Studio can see the resources you expect to allocate.

Next Step

After GPU Dashboard confirms the expected nodes and capacity:

1. Choose a platform model in Model Gallery or upload your own asset in My Models.
2. Follow Create Deployment.
3. Review Usage and Billing after the workload starts.