Azure OpenAI integration

AI: Azure-hosted OpenAI – chat completions, embeddings, image generation

Help CentreConnectors

Overview

Azure OpenAI is Microsoft's hosted OpenAI service – same models (GPT-4, GPT-3.5, DALL-E, embeddings) as openai.com but on Azure, with enterprise data residency and SLAs. The Automize Azure OpenAI connector covers chat completions, embeddings, image generation, and deployment listing.

  • Regions: Models are region-specific – not every region has every model. Common: East US, Sweden Central, France Central for newest models. Pick the region matching your Azure OpenAI resource on the connection.
  • Plans: Azure OpenAI requires manual approval – apply at aka.ms/oai/access. Pay-as-you-go after approval, with provisioned throughput options.

Setting up the connection

Auth type: API Key.

  1. Create an Azure OpenAI resource in Azure Portal (requires application approval).
  2. Deploy a model (e.g. gpt-4o, text-embedding-3-large) – note the deployment name (your alias for the model).
  3. Resource → Keys and Endpoint → copy KEY 1 + Endpoint URL.
  4. In Automize, open Settings → Connectors → Azure OpenAI → Add connection. Paste endpoint + API key.

Find your credentials at https://portal.azure.com/.


Rate limits

Token-based + request-based. Each deployment has TPM (tokens per minute) + RPM (requests per minute) – set when deploying. Connector retries 429 with Retry-After. For high throughput buy Provisioned Throughput Units (PTUs) – guaranteed capacity, no shared queue.


Data model

Azure OpenAI organises by deployments:

    Resource (regional) ─── Deployments ─── Model (gpt-4o, gpt-3.5-turbo, etc.)

You call deployments by name, not models. Same deployment name in one
region maps to a specific model. Reuse names across resources for
consistent calling code.

Available operations

4 operations available. Click any row to jump to its detail.

OperationCategoryWhat it does
Chat CompletionGeneralGenerate a chat completion
Create EmbeddingGeneralGenerate text embeddings
Generate ImageGeneralGenerate an image (DALL-E)
List DeploymentsGeneralList model deployments

Operations

Chat Completion

Generate a chat completion

ParameterRequiredTypeDescription
Deployment deploymentYestext
Messages messagesYestextarea
Temperature temperatureYesnumber
Max Tokens max_tokensYesnumber
Tools toolsYestext

Tips

  • deployment: your deployment name (NOT the OpenAI model name).
  • messages: array of {role: 'system'|'user'|'assistant', content: '...'}.
  • tools: function-calling schemas; the model can call them and you handle the results.
  • temperature 0-2 (0 = deterministic). top_p alt to temperature.

Errors

  • DeploymentNotFound – deployment name mismatch – check Azure Portal → your resource → Deployments.
  • RateLimitExceeded – TPM or RPM hit. Either raise deployment limits, buy PTUs, or back off.

Create Embedding

Generate text embeddings

ParameterRequiredTypeDescription
Deployment deploymentYestext
Input inputYestext

Tips

  • deployment: an embedding-model deployment (e.g. text-embedding-3-large).
  • input: string or array of strings. Returns vector(s).
  • Batch up to 16 inputs per call for efficiency.

Generate Image

Generate an image (DALL-E)

ParameterRequiredTypeDescription
Deployment deploymentYestext
Prompt promptYestext
Size sizeYestext
N nYestext

Tips

  • deployment: a DALL-E deployment (dall-e-3 typical).
  • size: '1024x1024', '1792x1024', '1024x1792' for DALL-E 3.
  • n: number of images (max 1 for DALL-E 3; up to 10 for DALL-E 2).

List Deployments

List model deployments

No input parameters.

Tips

  • Returns deployments in your resource – names + underlying models. Useful for runtime discovery.

FAQ

Azure OpenAI vs OpenAI direct – when?
Azure: enterprise data residency (no data used for training, region-pinned), Microsoft SLAs, integrates with Entra ID + Private Endpoint. OpenAI direct: simpler signup, sometimes newer models faster. Use Azure for enterprise compliance; OpenAI for prototyping.
Why does my deployment name not work?
Deployment names are case-sensitive + region-specific. The same name in Sweden Central isn't the same in East US. Verify in Azure Portal → your resource → Deployments.
PTU (Provisioned Throughput Units) – when?
Buy PTUs when you need guaranteed latency (no shared-pool throttling) – usually at >10k req/day. Pay per unit per hour, regardless of usage.

Related connectors

See it working on your own data

Everything documented here ships with the platform – try the document tools free, or go live in 7 days.