Skip to main content

AI Configuration

Aicser Enterprise Edition uses Azure OpenAI for natural language queries and AI chart generation.

Create an Azure OpenAI resource​

  1. Log in to the Azure portal
  2. Create an Azure OpenAI resource in your subscription
  3. Go to Model deployments and deploy a model (GPT-4.1-mini or GPT-4o-mini recommended)
  4. Note the:
    • API key (from Keys and Endpoint)
    • Endpoint URL (from Keys and Endpoint)
    • API version (use 2024-02-15-preview or latest stable)
    • Deployment name (the name you gave the deployment)

Configure environment variables​

In deploy/.env:

AZURE_OPENAI_API_KEY=your-api-key
AZURE_OPENAI_ENDPOINT=https://your-resource-name.openai.azure.com/
AZURE_OPENAI_API_VERSION=2024-02-15-preview
AZURE_OPENAI_DEPLOYMENT_NAME=your-deployment-name

The endpoint must end with a /.

Two-model setup (optional)​

EE supports routing between a primary and secondary model:

# Primary model — used for most queries
AZURE_OPENAI_DEPLOYMENT_NAME=gpt-4.1-mini

# Secondary model — used for complex reasoning
AZURE_OPENAI_GPT41_API_KEY=your-api-key
AZURE_OPENAI_GPT41_ENDPOINT=https://your-resource.openai.azure.com/
AZURE_OPENAI_GPT41_API_VERSION=2024-02-15-preview
AZURE_OPENAI_GPT41_DEPLOYMENT_NAME=gpt-4.1

If only the primary model is configured, EE uses it for all queries.

Verify​

After starting EE, try a natural language question in a chart. If AI is not working:

docker compose -f docker-compose.ee.yml logs server | grep -i "openai\|azure"

Common errors:

ErrorCause
AuthenticationErrorAZURE_OPENAI_API_KEY is wrong
DeploymentNotFoundAZURE_OPENAI_DEPLOYMENT_NAME does not match your Azure deployment
ResourceNotFoundAZURE_OPENAI_ENDPOINT URL format is wrong — must end with /