Overview
A client-mode connector for a curated slice of Azure OpenAI / Azure AI Foundry focused on IIoT / edge alarm text into Azure-hosted LLMs (OpenAI-compatible bodies, deployment-based paths):
Default
BASE_ADDRESS placeholder is https://YOUR_RESOURCE.openai.azure.com (no trailing slash) — replace with your real resource host. Default api-version in docs/examples: 2024-10-21. Topics use ai/azure-openai/<object>/<verb>.
See the base guide: REST API Connectors.
Auth:
api-key: <Azure OpenAI resource key> on every request (template default). Prefer Hub secrets (GET SECRET "AZURE_OPENAI_API_KEY" or AZURE_OPENAI_KEY) mapped into MQTT api_token. Entra ID alternative: Authorization: Bearer <token> with scope https://cognitiveservices.azure.com/.default. Never commit real keys.Factory live smoke is blocked until a resource endpoint + API key + deployment name are supplied. Set
BASE_ADDRESS to your https://{resource}.openai.azure.com host before production use.Basic Syntax
AzureOpenaiAndAzureAiFoundryRestApi.lot in the factory package (no ADD METADATA).
Topics
Configuration
- Set
BASE_ADDRESStohttps://{your-resource}.openai.azure.com(no trailing slash). - Store the Azure OpenAI API key in Hub secrets (
AZURE_OPENAI_API_KEYorAZURE_OPENAI_KEY); map into MQTTapi_token. - Pass
deployment(deployment name) andapi_version(e.g.2024-10-21) on inference MQTT JSON. - For alarm classification: publish to
ai/azure-openai/chat/completions/postwith an OpenAI-compatible messages body (optionalresponse_format). - For semantic search / similarity: publish to
ai/azure-openai/embeddings/postwith a tinyinputstring against an embeddings deployment. - Use
GetModelsas a cheap auth smoke before production workloads.

