`azure://`¶

Azure OpenAI Service — OpenAI's model line-up hosted on Azure with IAM, regional residency, private networking, and per-deployment naming. The wire format is OpenAI-compatible but the URL pattern and auth header differ from openai://, so Azure has its own scheme.

The key Azure-specific quirks this component handles:

URL: {baseURL}/openai/deployments/{model}/chat/completions?api-version={apiVersion}
Auth header: api-key: <key> (not Authorization: Bearer <key>).
baseURL is required — there is no default Azure endpoint.

URI shape¶

azure://chat?model=<deployment>&apiKey=<key>&baseURL=https://<resource>.openai.azure.com&apiVersion=<version>

The model parameter on Azure refers to the deployment name you chose in the Azure portal, not the model family — e.g. you might deploy GPT-4o under the deployment name prod-gpt4o.

Parameters¶

Param	Type	Default	Producer/Consumer/Both	Description
`model`	string	`gpt-4o`	Producer	Azure deployment name (NOT the OpenAI model id).
`apiKey`	string	(from constructor)	Producer	Azure OpenAI resource key. Sent as `api-key` header.
`baseURL`	string	required, no default	Producer	Your Azure resource endpoint, e.g. `https://my-resource.openai.azure.com`. URI creation fails if unset.
`apiVersion`	string	`2024-10-21`	Producer	Azure REST API version. Pin to a supported version.
`temperature`	float	`0.7`	Producer	Sampling temperature.
`maxTokens`	int	`4096`	Producer	Output cap.

Operation defaults to chat; embeddings is not supported on this scheme (use a deployment of text-embedding-3-* via the OpenAI scheme with baseURL pointed at Azure, or wait for the Azure embeddings code path).

Headers¶

Reads:

Header	Effect
`systemPrompt`	Prepended as a `role: system` message.

Sets:

Header	Description
`azure.model`	The deployment name used.
`azure.id`	OpenAI-shaped completion id.
`azure.finishReason`	`stop`, `length`, `content_filter`, etc.
`azure.usage.promptTokens` / `completionTokens` / `totalTokens`	Token billing.

Examples¶

- id: azure-chat
  from: direct://chat
  steps:
    - to: azure://chat?model=prod-gpt4o&baseURL=https://my-resource.openai.azure.com&apiKey=${AZURE_OPENAI_KEY}

Pin a specific API version (recommended for production stability):

- to: azure://chat?model=prod-gpt4o&baseURL=https://my-resource.openai.azure.com&apiVersion=2024-10-21&apiKey=${AZURE_OPENAI_KEY}

Behavior notes¶

The Azure response shape is OpenAI-compatible — the component uses llmutil.ChatRequest/ChatResponse/APIError types directly.
baseURL is required. Helix rejects the URI at creation time if unset; there's no "default Azure endpoint".
Azure AD / managed-identity auth is not supported on this scheme — only key-based auth. For AAD-token auth, run the route on a host inside the Azure VNet and bridge via an Azure API Management policy that swaps tokens for keys.
No embeddings, no streaming, no DALL-E on this scheme.
120s timeout. No retries.

Source¶

pkg/component/azure/azure.go
pkg/component/azure/azure_test.go

azure://¶