OCR

Once an OCR provider is configured, AI experts can extract text from images and PDFs — turning scanned documents, photos, and screenshots into editable text or Markdown.

![Settings — OCR](https://cdn.yaoagents.cn/assets/en/ocr/01-ocr-settings-en-$__THEME.png)

## Supported Providers

| Provider | Type | Description |
|----------|------|-------------|
| [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR) | Self-hosted | Open-source OCR engine, 100+ languages, runs on CPU or GPU. |
| [Baidu OCR](https://cloud.baidu.com/product/ocr) | Cloud | General and high-accuracy text recognition. |
| [Google Cloud Vision](https://cloud.google.com/vision) | Cloud | Best multilingual coverage, supports GenAI custom extraction. |
| [Azure Document Intelligence](https://azure.microsoft.com/products/ai-services/ai-document-intelligence) | Cloud | Strong at form and key-value extraction, supports hidden text layers. |

You only need one. Pick whichever suits your needs.

## Setup

1. Sign up with your chosen provider and get an API key.
2. Open **Settings → OCR**.
3. Select the provider and enter your credentials.
4. Click **Save**.

Required fields per provider:

| Provider | Fields |
|----------|--------|
| PaddleOCR | Service URL (your PaddleOCR Serving endpoint), API Key (optional) |
| Baidu OCR | API Key, Secret Key |
| Google Cloud Vision | API Key |
| Azure Document Intelligence | API Key, Endpoint |

## Recognition Types

| Type | Description |
|------|-------------|
| `general` | General text recognition (default) |
| `table` | Table extraction |
| `handwriting` | Handwritten text |
| `document` | Document layout analysis |
| `invoice` | Invoices |
| `receipt` | Receipts |
| `id_card` | ID cards |
| `passport` | Passports |
| `bank_card` | Bank cards |
| `license` | Business licenses |
| `license_plate` | License plates |

Coverage varies by provider. Unsupported types automatically fall back to `general`.

## Usage

After setup, OCR kicks in automatically when you send an attachment to an AI expert. Common examples:

- Upload a scanned PDF and ask the expert to extract text as Markdown
- Snap a photo of handwritten notes and have the expert transcribe them
- Upload an invoice or receipt for the expert to pull out key fields

![OCR Task — PDF to Markdown](https://cdn.yaoagents.cn/assets/en/ocr/02-ocr-task-en-$__THEME.png)