uniocr 0.2.3

Unified multilingual OCR tool based on PaddleOCR-VL and Apple Vision.

One API. Multiple engines. Zero friction.

---

UniOCR is a unified, multilingual OCR abstraction layer that wraps best-in-class OCR engines behind a single, clean interface. Throw any image or PDF at it — get back structured text, Markdown, and layout blocks — regardless of which engine runs under the hood.

Built for developers, AI agents, and automation pipelines (n8n, Dify, Telegram bots, etc.).

✨ Highlights

• 🔌 Pluggable engines — PaddleOCR-VL (deep document AI) and Apple Vision (native macOS) with automatic priority fallback
• ⚡ Zero-config acceleration — Auto-detects Apple Silicon → launches MLX-VLM → Neural Engine speedup. No manual setup.
• 📄 Accepts Anything — File paths, URLs, Base64, multi-page PDFs (auto-flattened).
• 📦 Unified Output — Supports .text, .markdown, .json, and Searchable Dual-Layer PDFs.
• 🌐 Built-in REST API — FastAPI powered, Swagger docs, batch processing — directly consumable by n8n / Dify / any HTTP client.
• 🐳 Docker ready — single command deployment via Docker Compose
• 🖥️ CLI — uniocr extract, uniocr engines, uniocr serve

🏗️ Architecture

┌──────────────────────────────────────────────────────────────┐
│                    User Interface Layer                       │
│              SDK  ·  CLI  ·  REST API                        │
├──────────────────────────────────────────────────────────────┤
│                    Input Processor                           │
│         URL → File  ·  PDF → Images  ·  Base64 → File       │
├──────────────────────────────────────────────────────────────┤
│                Engine Dispatcher (auto)                      │
│         PaddleOCR-VL → Apple Vision → fallback               │
├─────────────────────┬────────────────────────────────────────┤
│   PaddleOCR-VL      │        Apple Vision                    │
│   + MLX-VLM         │        (native macOS)                  │
│   (auto-accelerated)│                                        │
├─────────────────────┴────────────────────────────────────────┤
│                 Standardised Output                          │
│          Document → Pages → Blocks                           │
│          .text  ·  .markdown  ·  .to_dict()                  │
└──────────────────────────────────────────────────────────────┘

🚀 Quick Start

Option 1: pip install

# Core only (lightweight, includes PDF flattening)
pip install uniocr

# With PaddleOCR-VL (powerful document AI, ~1.8 GB model download on first run)
pip install "uniocr[paddle]"

# With Apple Vision (macOS only, uses built-in system OCR)
pip install "uniocr[apple]"

# With REST API server
pip install "uniocr[api]"

# Everything
pip install "uniocr[all]"

Option 2: Docker (recommended for servers)

# Quick start — pull and run in detached mode
docker run -d --name uniocr -p 8000:8000 ghcr.io/yuanweize/uni-ocr:latest

# Or use Docker Compose (recommended)
curl -O https://raw.githubusercontent.com/yuanweize/uni-ocr/master/docker-compose.yml
docker compose up -d

# Check it's running
curl http://localhost:8000/health

📖 Usage

Python SDK

from uniocr import UniOCR

ocr = UniOCR(engine="auto")          # Auto-selects best available engine
doc = ocr.extract("invoice.pdf")

print(doc.text)                       # Plain text
print(doc.markdown)                   # Structured Markdown
print(doc.to_dict())                  # JSON-serialisable dict

# Access individual blocks with layout info
for page in doc.pages:
    for block in page.blocks:
        print(f"[{block.block_type}] {block.text}")
        print(f"  bbox: {block.bbox}, confidence: {block.confidence}")

CLI

# List available engines
uniocr engines
# Output:
#   Available engines:
#     • paddle
#     • apple

# Extract text (outputs Markdown by default)
uniocr extract document.pdf -o result.md

# Generate a Searchable PDF (automatically triggered by .pdf extension)
uniocr extract input_image.jpg -o output_searchable.pdf

# Specify engine and output format
uniocr extract scan.png --engine apple --format json -o result.json

# Extract from a URL
uniocr extract "https://example.com/receipt.png" --format text

# Start the API server (single worker)
uniocr serve --port 8000

# Production: multiple workers
uniocr serve --port 8000 --workers 4

REST API

Start the server:

uniocr serve --port 8000
# Or via Docker:
docker compose up -d

Endpoints

Method	Endpoint	Description
GET	/health	Health check & engine list
GET	/engines	List available OCR engines
GET	/docs	Interactive Swagger docs
POST	/extract	Extract text from uploaded file (JSON/Markdown)
POST	/extract/pdf	Extract text and return a Searchable PDF file
POST	/extract/url	Extract text from URL
POST	/extract/batch	Process multiple files

Examples

# Health check
curl http://localhost:8000/health
# → {"status":"ok","version":"0.2.2","engines":["paddle","apple"]}

# Upload a file
curl -X POST http://localhost:8000/extract \
  -F "file=@invoice.pdf" \
  -F "engine=auto"

# Extract via URL
curl -X POST http://localhost:8000/extract/url \
  -F "url=https://example.com/image.png"

# Return a Searchable PDF directly
curl -X POST http://localhost:8000/extract/pdf \
  -F "file=@scan.png" -o searchable.pdf

# Batch processing
curl -X POST http://localhost:8000/extract/batch \
  -F "files=@page1.png" -F "files=@page2.png" \
  -F "engine=auto"

Response format

{
  "request_id": "ab07767c-331f-4f26-be01-2fcb75d36149",
  "engine": "PaddleOCRVLAdapter",
  "page_count": 1,
  "text": "Invoice #12345\nTotal: €1,234.56",
  "markdown": "# Invoice #12345\n\nTotal: €1,234.56",
  "pages": [
    {
      "page_number": 1,
      "text": "...",
      "markdown": "...",
      "blocks": [
        {
          "block_type": "text",
          "text": "Invoice #12345",
          "bbox": [0.05, 0.02, 0.45, 0.06],
          "confidence": 0.98
        }
      ]
    }
  ],
  "elapsed_seconds": 2.35
}

🐳 Docker

Quick Run

# Run in background (detached mode)
docker run -d \
  --name uniocr \
  -p 8000:8000 \
  -v uniocr-models:/root/.paddlex \
  ghcr.io/yuanweize/uni-ocr:latest

Docker Compose (recommended)

# Download compose file
curl -O https://raw.githubusercontent.com/yuanweize/uni-ocr/master/docker-compose.yml

# Start in background
docker compose up -d

# View logs
docker compose logs -f

# Stop
docker compose down

Build locally

git clone https://github.com/yuanweize/uni-ocr.git
cd uni-ocr
docker compose up -d --build

🔧 Engine Priority

When engine="auto", UniOCR selects the best available engine:

Priority	Engine	Best for	Speed
1	PaddleOCR-VL + MLX-VLM	Complex layouts, tables, formulas, 109 languages	⚡⚡
2	PaddleOCR-VL (CPU)	Same capabilities, without MLX acceleration	⚡
3	Apple Vision	Simple text, macOS only, instant	⚡⚡⚡

Apple Silicon users: when mlx-vlm is installed, UniOCR automatically starts an MLX-VLM server for Neural Engine acceleration. No configuration needed. The server is cleaned up on exit.

🔗 Integration Examples

UniOCR is designed to be called by automation tools and AI agents.

n8n Workflow

Use the HTTP Request node to call UniOCR:

Telegram Trigger → HTTP Request (UniOCR /extract) → AI Agent → ERPNext API

Configuration:

• Method: POST
• URL: http://uniocr:8000/extract
• Body: Form-Data, file = {{ $binary.data }}

Dify Tool

Add UniOCR as a custom tool in Dify with the OpenAPI spec at /docs.

Bob (macOS OCR Plugin)

UniOCR can serve as the OCR backend for Bob:

# Start UniOCR on the default port
uniocr serve --port 8000
# Bob → Preferences → OCR → Custom API → http://localhost:8000/extract

Shell / Scripts

# Quick OCR from clipboard image (macOS)
pbpaste | base64 | curl -s -X POST http://localhost:8000/extract \
  -F "file=@-;filename=clipboard.png" | jq .text

⚙️ Configuration

UniOCR works out of the box with zero configuration. For advanced use cases:

Environment Variable	Description	Default
UNIOCR_PORT	API server port (Docker Compose)	8000
UNIOCR_MLX_VLM_URL	Override MLX-VLM server URL	Auto-detected
UNIOCR_MLX_VLM_MODEL	MLX-VLM model identifier	PaddlePaddle/PaddleOCR-VL-1.6

Copy .env.example to .env to customise:

cp .env.example .env

📁 Project Structure

uni-ocr/
├── src/uniocr/
│   ├── __init__.py          # UniOCR main class & public API
│   ├── models.py            # Document / Page / Block dataclasses
│   ├── cli.py               # CLI: extract · engines · serve
│   ├── api.py               # FastAPI REST service
│   ├── engines/
│   │   ├── __init__.py      # Engine registry & auto-dispatcher
│   │   ├── base.py          # BaseOCREngine ABC
│   │   ├── apple_vision.py  # macOS Vision adapter
│   │   └── paddle.py        # PaddleOCR-VL + MLX-VLM adapter
│   └── processors/
│       └── input.py         # URL / Base64 / PDF normalisation
├── assets/
│   └── logo.svg             # Project logo
├── Dockerfile
├── docker-compose.yml
├── .env.example
├── pyproject.toml
├── CLAUDE.md                # Development guidelines
├── LICENSE                  # MIT
├── README.md                # English docs (this file)
└── README_zh.md             # 中文文档

🤝 Contributing

Contributions are welcome! Please open an issue or pull request.

1. Fork the repo
2. Create a feature branch (git checkout -b feat/amazing-feature)
3. Commit your changes (git commit -m 'feat: add amazing feature')
4. Push to the branch (git push origin feat/amazing-feature)
5. Open a Pull Request

📄 License

MIT © 2026 Weize Yuan