Azure OpenAI Provider

Setup

Appam's Azure support uses the OpenAI Responses client with an Azure-specific provider enum.

export AZURE_OPENAI_API_KEY="..."
export AZURE_OPENAI_RESOURCE="my-resource"
export AZURE_OPENAI_API_VERSION="2025-04-01-preview"   # optional
export AZURE_OPENAI_MODEL="gpt-5.4"                    # optional, used by the example binary

The resource name is the subdomain of your Azure endpoint. In the current crate, Azure URLs are built as:

https://{resource}.cognitiveservices.azure.com/openai/responses?api-version={version}

Construction

Agent::quick() does not auto-detect Azure. Use AgentBuilder with LlmProvider::AzureOpenAI:

use appam::prelude::*;

let agent = AgentBuilder::new("azure-agent")
    .provider(LlmProvider::AzureOpenAI {
        resource_name: "my-resource".to_string(),
        api_version: "2025-04-01-preview".to_string(),
    })
    .model("gpt-5.4")
    .system_prompt("You are a helpful assistant.")
    .build()?;

Authentication

Azure requests use the api-key header. The client resolves credentials in this order:

config.api_key
AZURE_OPENAI_API_KEY
OPENAI_API_KEY

Reasoning

Azure reuses the same OpenAI reasoning configuration used by the direct OpenAI provider:

use appam::prelude::*;

let agent = AgentBuilder::new("azure-reasoning")
    .provider(LlmProvider::AzureOpenAI {
        resource_name: "my-resource".to_string(),
        api_version: "2025-04-01-preview".to_string(),
    })
    .model("gpt-5.4")
    .system_prompt("You are a reasoning assistant.")
    .openai_reasoning(ReasoningConfig::high_effort())
    .build()?;

Deployment Names

.model(...) should match the Azure deployment name you created. The example binary in this crate also reads AZURE_OPENAI_MODEL for that value.

Example Binary

The crate includes a working Azure sample:

cargo run --example coding-agent-azure-openai