Azure AI Document Intelligence Ts

1---
2name: azure-ai-document-intelligence-ts
3description: Extract text, tables, and structured data from documents using Azure Document Intelligence (@azure-rest/ai-document-intelligence). Use when processing invoices, receipts, IDs, forms, or building custom document models.
4package: "@azure-rest/ai-document-intelligence"
5---
6 
7# Azure Document Intelligence REST SDK for TypeScript
8 
9Extract text, tables, and structured data from documents using prebuilt and custom models.
10 
11## Installation
12 
13```bash
14npm install @azure-rest/ai-document-intelligence @azure/identity
15```
16 
17## Environment Variables
18 
19```bash
20DOCUMENT_INTELLIGENCE_ENDPOINT=https://<resource>.cognitiveservices.azure.com
21DOCUMENT_INTELLIGENCE_API_KEY=<api-key>
22```
23 
24## Authentication
25 
26**Important**: This is a REST client. `DocumentIntelligence` is a **function**, not a class.
27 
28### DefaultAzureCredential
29 
30```typescript
31import DocumentIntelligence from "@azure-rest/ai-document-intelligence";
32import { DefaultAzureCredential } from "@azure/identity";
33 
34const client = DocumentIntelligence(
35  process.env.DOCUMENT_INTELLIGENCE_ENDPOINT!,
36  new DefaultAzureCredential()
37);
38```
39 
40### API Key
41 
42```typescript
43import DocumentIntelligence from "@azure-rest/ai-document-intelligence";
44 
45const client = DocumentIntelligence(
46  process.env.DOCUMENT_INTELLIGENCE_ENDPOINT!,
47  { key: process.env.DOCUMENT_INTELLIGENCE_API_KEY! }
48);
49```
50 
51## Analyze Document (URL)
52 
53```typescript
54import DocumentIntelligence, {
55  isUnexpected,
56  getLongRunningPoller,
57  AnalyzeOperationOutput
58} from "@azure-rest/ai-document-intelligence";
59 
60const initialResponse = await client
61  .path("/documentModels/{modelId}:analyze", "prebuilt-layout")
62  .post({
63    contentType: "application/json",
64    body: {
65      urlSource: "https://example.com/document.pdf"
66    },
67    queryParameters: { locale: "en-US" }
68  });
69 
70if (isUnexpected(initialResponse)) {
71  throw initialResponse.body.error;
72}
73 
74const poller = getLongRunningPoller(client, initialResponse);
75const result = (await poller.pollUntilDone()).body as AnalyzeOperationOutput;
76 
77console.log("Pages:", result.analyzeResult?.pages?.length);
78console.log("Tables:", result.analyzeResult?.tables?.length);
79```
80 
81## Analyze Document (Local File)
82 
83```typescript
84import { readFile } from "node:fs/promises";
85 
86const fileBuffer = await readFile("./document.pdf");
87const base64Source = fileBuffer.toString("base64");
88 
89const initialResponse = await client
90  .path("/documentModels/{modelId}:analyze", "prebuilt-invoice")
91  .post({
92    contentType: "application/json",
93    body: { base64Source }
94  });
95 
96if (isUnexpected(initialResponse)) {
97  throw initialResponse.body.error;
98}
99 
100const poller = getLongRunningPoller(client, initialResponse);
101const result = (await poller.pollUntilDone()).body as AnalyzeOperationOutput;
102```
103 
104## Prebuilt Models
105 
106| Model ID | Description |
107|----------|-------------|
108| `prebuilt-read` | OCR - text and language extraction |
109| `prebuilt-layout` | Text, tables, selection marks, structure |
110| `prebuilt-invoice` | Invoice fields |
111| `prebuilt-receipt` | Receipt fields |
112| `prebuilt-idDocument` | ID document fields |
113| `prebuilt-tax.us.w2` | W-2 tax form fields |
114| `prebuilt-healthInsuranceCard.us` | Health insurance card fields |
115| `prebuilt-contract` | Contract fields |
116| `prebuilt-bankStatement.us` | Bank statement fields |
117 
118## Extract Invoice Fields
119 
120```typescript
121const initialResponse = await client
122  .path("/documentModels/{modelId}:analyze", "prebuilt-invoice")
123  .post({
124    contentType: "application/json",
125    body: { urlSource: invoiceUrl }
126  });
127 
128if (isUnexpected(initialResponse)) {
129  throw initialResponse.body.error;
130}
131 
132const poller = getLongRunningPoller(client, initialResponse);
133const result = (await poller.pollUntilDone()).body as AnalyzeOperationOutput;
134 
135const invoice = result.analyzeResult?.documents?.[0];
136if (invoice) {
137  console.log("Vendor:", invoice.fields?.VendorName?.content);
138  console.log("Total:", invoice.fields?.InvoiceTotal?.content);
139  console.log("Due Date:", invoice.fields?.DueDate?.content);
140}
141```
142 
143## Extract Receipt Fields
144 
145```typescript
146const initialResponse = await client
147  .path("/documentModels/{modelId}:analyze", "prebuilt-receipt")
148  .post({
149    contentType: "application/json",
150    body: { urlSource: receiptUrl }
151  });
152 
153const poller = getLongRunningPoller(client, initialResponse);
154const result = (await poller.pollUntilDone()).body as AnalyzeOperationOutput;
155 
156const receipt = result.analyzeResult?.documents?.[0];
157if (receipt) {
158  console.log("Merchant:", receipt.fields?.MerchantName?.content);
159  console.log("Total:", receipt.fields?.Total?.content);
160  
161  for (const item of receipt.fields?.Items?.values || []) {
162    console.log("Item:", item.properties?.Description?.content);
163    console.log("Price:", item.properties?.TotalPrice?.content);
164  }
165}
166```
167 
168## List Document Models
169 
170```typescript
171import DocumentIntelligence, { isUnexpected, paginate } from "@azure-rest/ai-document-intelligence";
172 
173const response = await client.path("/documentModels").get();
174 
175if (isUnexpected(response)) {
176  throw response.body.error;
177}
178 
179for await (const model of paginate(client, response)) {
180  console.log(model.modelId);
181}
182```
183 
184## Build Custom Model
185 
186```typescript
187const initialResponse = await client.path("/documentModels:build").post({
188  body: {
189    modelId: "my-custom-model",
190    description: "Custom model for purchase orders",
191    buildMode: "template",  // or "neural"
192    azureBlobSource: {
193      containerUrl: process.env.TRAINING_CONTAINER_SAS_URL!,
194      prefix: "training-data/"
195    }
196  }
197});
198 
199if (isUnexpected(initialResponse)) {
200  throw initialResponse.body.error;
201}
202 
203const poller = getLongRunningPoller(client, initialResponse);
204const result = await poller.pollUntilDone();
205console.log("Model built:", result.body);
206```
207 
208## Build Document Classifier
209 
210```typescript
211import { DocumentClassifierBuildOperationDetailsOutput } from "@azure-rest/ai-document-intelligence";
212 
213const containerSasUrl = process.env.TRAINING_CONTAINER_SAS_URL!;
214 
215const initialResponse = await client.path("/documentClassifiers:build").post({
216  body: {
217    classifierId: "my-classifier",
218    description: "Invoice vs Receipt classifier",
219    docTypes: {
220      invoices: {
221        azureBlobSource: { containerUrl: containerSasUrl, prefix: "invoices/" }
222      },
223      receipts: {
224        azureBlobSource: { containerUrl: containerSasUrl, prefix: "receipts/" }
225      }
226    }
227  }
228});
229 
230if (isUnexpected(initialResponse)) {
231  throw initialResponse.body.error;
232}
233 
234const poller = getLongRunningPoller(client, initialResponse);
235const result = (await poller.pollUntilDone()).body as DocumentClassifierBuildOperationDetailsOutput;
236console.log("Classifier:", result.result?.classifierId);
237```
238 
239## Classify Document
240 
241```typescript
242const initialResponse = await client
243  .path("/documentClassifiers/{classifierId}:analyze", "my-classifier")
244  .post({
245    contentType: "application/json",
246    body: { urlSource: documentUrl },
247    queryParameters: { split: "auto" }
248  });
249 
250if (isUnexpected(initialResponse)) {
251  throw initialResponse.body.error;
252}
253 
254const poller = getLongRunningPoller(client, initialResponse);
255const result = await poller.pollUntilDone();
256console.log("Classification:", result.body.analyzeResult?.documents);
257```
258 
259## Get Service Info
260 
261```typescript
262const response = await client.path("/info").get();
263 
264if (isUnexpected(response)) {
265  throw response.body.error;
266}
267 
268console.log("Custom model limit:", response.body.customDocumentModels.limit);
269console.log("Custom model count:", response.body.customDocumentModels.count);
270```
271 
272## Polling Pattern
273 
274```typescript
275import DocumentIntelligence, {
276  isUnexpected,
277  getLongRunningPoller,
278  AnalyzeOperationOutput
279} from "@azure-rest/ai-document-intelligence";
280 
281// 1. Start operation
282const initialResponse = await client
283  .path("/documentModels/{modelId}:analyze", "prebuilt-layout")
284  .post({ contentType: "application/json", body: { urlSource } });
285 
286// 2. Check for errors
287if (isUnexpected(initialResponse)) {
288  throw initialResponse.body.error;
289}
290 
291// 3. Create poller
292const poller = getLongRunningPoller(client, initialResponse);
293 
294// 4. Optional: Monitor progress
295poller.onProgress((state) => {
296  console.log("Status:", state.status);
297});
298 
299// 5. Wait for completion
300const result = (await poller.pollUntilDone()).body as AnalyzeOperationOutput;
301```
302 
303## Key Types
304 
305```typescript
306import DocumentIntelligence, {
307  isUnexpected,
308  getLongRunningPoller,
309  paginate,
310  parseResultIdFromResponse,
311  AnalyzeOperationOutput,
312  DocumentClassifierBuildOperationDetailsOutput
313} from "@azure-rest/ai-document-intelligence";
314```
315 
316## Best Practices
317 
3181. **Use getLongRunningPoller()** - Document analysis is async, always poll for results
3192. **Check isUnexpected()** - Type guard for proper error handling
3203. **Choose the right model** - Use prebuilt models when possible, custom for specialized docs
3214. **Handle confidence scores** - Fields have confidence values, set thresholds for your use case
3225. **Use pagination** - Use `paginate()` helper for listing models
3236. **Prefer neural mode** - For custom models, neural handles more variation than template
324