Bolt.new AI and Azure Machine Learning integration: Step-by-Step Guide 2025

To integrate Bolt.new with Azure Machine Learning, you don’t “connect” Bolt itself to Azure; instead, you write normal client code inside your Bolt workspace that calls Azure ML’s REST endpoints or uses the official Azure ML REST API to submit jobs, call deployed models, or manage pipelines. Bolt acts as the environment where you write and run this integration code, while Azure ML stays the external system. You authenticate using Azure AD (typically a Service Principal) and call your Azure ML workspace endpoints from Bolt using standard HTTP requests.

 

What You Actually Do

 

You create Azure Machine Learning resources on Azure (workspace + model deployment or online endpoint), then in Bolt.new you:

• Set environment variables for Azure credentials and workspace details.
• Use a Service Principal for secure, non-interactive auth (client ID, tenant ID, client secret).
• Request an Azure AD access token using OAuth 2.0 client credentials.
• Call your Azure ML endpoint through its REST API with that token.

That’s the entire integration pattern: Bolt → REST → Azure Machine Learning.

 

Step-by-step Integration

 

This is the simplest stable pattern used in real systems.

• Create a Service Principal in Azure Portal and assign it access to your Azure ML Workspace.
• Collect required values:
  • AZURE_TENANT_ID
  • AZURE_CLIENT_ID
  • AZURE_CLIENT_SECRET
  • AZURE_ML_ENDPOINT\_URL (this is your deployed model's scoring endpoint URL)
• Configure environment variables inside Bolt.new’s environment panel.
• Write API-call code inside Bolt to authenticate and call your Azure ML endpoint.

 

Working Code Example (Node.js)

 

This version uses pure REST calls (no Azure SDK required) because it works in any environment including Bolt’s sandbox.

// This function gets an Azure AD token using OAuth2 client credentials
async function getAzureToken() {
  const tenantId = process.env.AZURE_TENANT_ID;
  const clientId = process.env.AZURE_CLIENT_ID;
  const clientSecret = process.env.AZURE_CLIENT_SECRET;

  const url = `https://login.microsoftonline.com/${tenantId}/oauth2/v2.0/token`;

  const params = new URLSearchParams();
  params.append("client_id", clientId);
  params.append("client_secret", clientSecret);
  params.append("grant_type", "client_credentials");
  params.append("scope", "https://management.azure.com/.default");

  const response = await fetch(url, {
    method: "POST",
    body: params
  });

  const data = await response.json();
  return data.access_token;
}

// Call a deployed Azure ML model endpoint
async function callAzureMLEndpoint(inputData) {
  const endpointUrl = process.env.AZURE_ML_ENDPOINT_URL; // example: https://xxxx.eastus.inference.ml.azure.com/score
  const token = await getAzureToken();

  const response = await fetch(endpointUrl, {
    method: "POST",
    headers: {
      "Authorization": `Bearer ${token}`,
      "Content-Type": "application/json"
    },
    body: JSON.stringify(inputData)
  });

  const result = await response.json();
  return result;
}

// Example usage inside Bolt
(async () => {
  const prediction = await callAzureMLEndpoint({
    input_data: [{ text: "hello world" }]
  });

  console.log("Azure ML Prediction:", prediction);
})();

 

Why This Works

 

Azure ML model deployments expose HTTPS endpoints. These are normal REST endpoints. Bolt.new can call any REST API if you supply credentials. The authentication layer is Azure AD, not Bolt. Bolt is simply running Node/JS code that fetches a token and calls your endpoint.

• No magic integration. You explicitly call Azure ML.
• Security is handled via Azure AD (Service Principal).
• Bolt only sees what you put in environment variables, so secrets stay isolated.

 

Once You Move to Production

 

After prototyping in Bolt.new, move your code into your real backend environment. Use a secrets manager (Azure Key Vault, AWS Secrets Manager, or environment variables) and ensure least-privilege Service Principal permissions. The REST call patterns remain identical.