Azure Anthropic Provider

Run Claude models through Azure-hosted Anthropic-compatible endpoints with Appam's Anthropic client.

Setup

Appam's Azure Anthropic support reuses the Anthropic Messages client with an Azure-specific provider enum.

export AZURE_API_KEY="..."  # or AZURE_ANTHROPIC_API_KEY
export AZURE_ANTHROPIC_BASE_URL="https://my-resource.services.ai.azure.com/anthropic"
# or export AZURE_ANTHROPIC_RESOURCE="my-resource"
export AZURE_ANTHROPIC_AUTH_METHOD="x_api_key"        # optional
export AZURE_ANTHROPIC_MODEL="claude-opus-4-6"        # optional, used by the example binary

The provider accepts a full base_url instead of hardcoding a single Azure host pattern. Both services.ai.azure.com and openai.azure.com Azure Anthropic hosts were validated against the same resource during implementation.

Construction

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

use appam::prelude::*;

let agent = AgentBuilder::new("azure-anthropic-agent")
    .provider(LlmProvider::AzureAnthropic {
        base_url: "https://my-resource.services.ai.azure.com/anthropic".to_string(),
        auth_method: appam::llm::anthropic::AzureAnthropicAuthMethod::XApiKey,
    })
    .model("claude-opus-4-6")
    .system_prompt("You are a helpful assistant.")
    .build()?;

Authentication

Azure Anthropic supports two auth modes:

AzureAnthropicAuthMethod::XApiKey
AzureAnthropicAuthMethod::BearerToken

The Anthropic client resolves credentials in this order:

config.api_key
Auth-specific Azure Anthropic env vars
AZURE_API_KEY

Anthropic Features

Azure Anthropic keeps the normal Anthropic Messages API body and streaming semantics, so the existing Anthropic builder surface still applies:

use appam::prelude::*;

let agent = AgentBuilder::new("azure-anthropic-thinking")
    .provider(LlmProvider::AzureAnthropic {
        base_url: "https://my-resource.services.ai.azure.com/anthropic".to_string(),
        auth_method: appam::llm::anthropic::AzureAnthropicAuthMethod::XApiKey,
    })
    .model("claude-opus-4-6")
    .system_prompt("You are a thoughtful assistant.")
    .thinking(appam::llm::anthropic::ThinkingConfig::adaptive())
    .effort(appam::llm::anthropic::EffortLevel::Max)
    .tool_choice(appam::llm::anthropic::ToolChoiceConfig::Auto {
        disable_parallel_tool_use: false,
    })
    .build()?;

Deployment Names

.model(...) should match the Azure deployment name you created. The example binary in this crate also reads AZURE_ANTHROPIC_MODEL for that value and defaults to claude-opus-4-6.

Example Binary

The crate includes a working Azure Anthropic sample:

cargo run --example coding-agent-azure-anthropic