Azure AI Document Intelligence Ts

Install

Quick Start

Auth: DefaultAzureCredential is for local development. See auth-best-practices.md for production patterns.

Non-Obvious Patterns

• REST client — DocumentIntelligence is a function, not a class
• Analyze path: client.path("/documentModels/{modelId}:analyze", "prebuilt-layout").post({...})
• Must use getLongRunningPoller(client, initialResponse) then poller.pollUntilDone()
• Local file: send as base64Source in body, not as binary stream
• Pagination: import { paginate } from "@azure-rest/ai-document-intelligence"

Best Practices

1. Use getLongRunningPoller() — document analysis is async, always poll
2. Check isUnexpected() — type guard for proper error handling
3. Choose the right model — prebuilt when possible, custom for specialized docs
4. Handle confidence scores — set thresholds for your use case
5. Use paginate() helper for listing models
6. Prefer neural mode for custom models over template