# AzureClaudeProvider

**Kind:** Class

**Source:** [`atloria-monorepo/apps/api/src/ai/providers/azure-claude.provider.ts`](https://github.com/sherkety/atloria/blob/main/atloria-monorepo/apps/api/src/ai/providers/azure-claude.provider.ts#L48)

Azure AI Foundry Claude Provider
Uses native Anthropic Messages API endpoint through Azure AI Foundry
Endpoint: /anthropic/v1/messages

Authentication priority:
1. API Key (ANTHROPIC_FOUNDRY_API_KEY) - simplest, no az login needed
2. Azure AD (DefaultAzureCredential/AzureCliCredential) - fallback

`AzureClaudeProvider` integrates Claude models hosted in Azure AI Foundry with the application's AI provider layer. It calls Azure's native Anthropic Messages API endpoint (`/anthropic/v1/messages`) for standard, usage-aware, and streaming text generation, then exposes analysis results through the shared provider interface. Authentication prefers `ANTHROPIC_FOUNDRY_API_KEY` and falls back to Azure AD credentials when no API key is configured.

**Implements:** `AIProvider`

## Methods

| Method | Signature | Returns |
|---|---|---|
| `getAzureAIConfig` | `getAzureAIConfig()` | `Promise<{ endpoint: string; accessToken?: string; apiKey?: string; modelName: string } | null>` |
| `isAvailable` | `isAvailable()` | `Promise<boolean>` |
| `generateText` | `generateText(prompt: string, options: GenerateOptions)` | `Promise<string>` |
| `generateTextWithUsage` | `generateTextWithUsage(prompt: string, options: GenerateOptions)` | `Promise<AIResponse>` |
| `generateStream` | `generateStream(prompt: string, options: GenerateOptions)` | `AsyncIterableIterator<string>` |
| `analyze` | `analyze(content: string, type: AnalysisType)` | `Promise<AnalysisResult>` |

## Properties

| Property | Type |
|---|---|
| `name` | `any` |

## Where it refuses work

- `AzureClaudeProvider` stops the work with `Error` when `!projectEndpoint` — “Azure AI Project configuration missing. Please set AZURE_EXISTING_AIPROJECT_ENDPOINT”.
- `AzureClaudeProvider` stops the work with an early return when `this.tokenCache && this.tokenCache.expiresAt > Date.now()`.
- `AzureClaudeProvider` stops the work with an early return when `this.apiKey`.
- `AzureClaudeProvider` stops the work with an early return when `data === '[DONE]'`.

## When something fails

- `AzureClaudeProvider` handles failure in 5 places: it turns it into a return value in 2, lets it reach the caller in 2, and logs it and continues in 1.

## Diagram

```mermaid
graph LR
  App[Application AI Request] --> Provider[AzureClaudeProvider]
  Provider --> Config[getAzureAIConfig]

  Config --> APIKey{ANTHROPIC_FOUNDRY_API_KEY available?}
  APIKey -->|Yes| KeyAuth[API Key Authentication]
  APIKey -->|No| AzureAD[DefaultAzureCredential / AzureCliCredential]

  KeyAuth --> Endpoint[Azure AI Foundry<br/>/anthropic/v1/messages]
  AzureAD --> Endpoint

  Endpoint --> Text[generateText]
  Endpoint --> Usage[generateTextWithUsage]
  Endpoint --> Stream[generateStream]
  Text --> Analysis[analyze]
  Usage --> Response[AIResponse]
  Stream --> Chunks[AsyncIterableIterator<string>]
```

## Usage

```ts
import { AzureClaudeProvider } from './ai/providers/azure-claude.provider';

const provider = new AzureClaudeProvider();

if (!(await provider.isAvailable())) {
  throw new Error(
    'Azure Claude is unavailable. Configure ANTHROPIC_FOUNDRY_API_KEY or Azure AD credentials.',
  );
}

const response = await provider.generateTextWithUsage(
  'Summarize the following incident report in three bullet points.',
  {
    systemPrompt: 'You are a concise engineering incident assistant.',
  },
);

console.log(response.content);
console.log('Token usage:', response.usage);

// Stream responses when incremental output is needed.
for await (const chunk of provider.generateStream('Explain this error log.')) {
  process.stdout.write(chunk);
}
```

## AI Coding Instructions

- Use `getAzureAIConfig()` as the single source of truth for endpoint, model, and authentication configuration; do not duplicate credential resolution in request methods.
- Prefer `ANTHROPIC_FOUNDRY_API_KEY` for local development and service deployments because it does not require an active Azure CLI session.
- Preserve Azure AD fallback behavior with `DefaultAzureCredential` or `AzureCliCredential` when the API key is absent.
- Send requests using the native Anthropic Messages API contract at `/anthropic/v1/messages`; do not convert requests to Azure OpenAI chat-completions format.
- Ensure streaming methods yield text chunks incrementally and that usage-aware methods return the shared `AIResponse` structure.

## Referenced By

- `AIModule` (MODULE_PROVIDES)
- `AIModule` (MODULE_EXPORTS)
- `AIService` (DEPENDS_ON)
- `BusinessContextParserService` (DEPENDS_ON)
- `DocumentationService` (DEPENDS_ON)
- `CompetitorDiscoveryService` (DEPENDS_ON)
- `DocAutomationService` (DEPENDS_ON)
- `FeatureDocsGeneratorService` (DEPENDS_ON)
- `Neo4jDocsOrchestratorService` (DEPENDS_ON)
- `TutorialGeneratorService` (DEPENDS_ON)
- `UserDocsGeneratorService` (DEPENDS_ON)
- `WorkflowDiscoveryService` (DEPENDS_ON)