DeepSeek OCR

Recognize text in images using the DeepSeek-OCR model.

Quick start

{baseDir}/scripts/ocr.sh /path/to/image.jpg

Usage

{baseDir}/scripts/ocr.sh <image_path> [output_format]

Parameters:

• : Local image file (jpg, png, webp, gif, bmp)
• [output_format]: Optional, defaults to markdown. Can be text, json, etc.

Examples

# Convert to markdown (default)
{baseDir}/scripts/ocr.sh /path/to/image.jpg

# Convert to plain text
{baseDir}/scripts/ocr.sh /path/to/image.png text

# Extract table as JSON
{baseDir}/scripts/ocr.sh /path/to/table.jpg "extract table as json"

Remote URL images

The model only supports base64-encoded images. For remote URLs, download first:

curl -s -o /tmp/image.jpg "https://example.com/image.jpg"
{baseDir}/scripts/ocr.sh /tmp/image.jpg

API key

Set DEEPSEEK_OCR_API_KEY, or configure in ~/.openclaw/openclaw.json:

{
  skills: {
    "deepseek-ocr": {
      apiKey: "YOUR_KEY_HERE",
    },
  },
}

Default API URL: https://api.modelverse.cn/v1/chat/completions

Override with DEEPSEEK_OCR_API_URL if needed.