Visibabel

Local OCR and translation for your browser. Visibabel captures images from web pages, runs OCR through GLM-OCR, and optionally augments results with layout data from a companion Python service.

Core modules

Module	Path	Role
Browser extension	extension/	MV3 Chrome extension for capture, OCR, translation, and results UI
GLM-OCR service	glm-ocr/	FastAPI service for layout-aware OCR augmentation

Shared dev tooling lives in ollama/ (launcher scripts and endpoint smoke tests).

Architecture

flowchart TD
  subgraph Browser extension
  pipeline[Request OCR]
  translat[Request translation]
  final[Final text]
  end

  subgraph "GLM-OCR SDK server (optional)"
  glmsdk[PP-DocLayoutV3 model]
  end

  subgraph Ollama server
  glmmodel[GLM-OCR model]
  transmodel[Translation model]
  end

  pipeline -->|requests OCR| glmmodel
  pipeline -->|"requests layout data"| glmsdk
  glmsdk -->|augments output| glmmodel
  glmmodel -->|provides captured text| translat
  translat -->|requests translation| transmodel
  transmodel -->|provides translated text| final

Loading

Prerequisites

• LLM server API running glm-ocr:latest
• Node.js 20+
• Python 3.12+
• Google Chrome for extension development

Quick start

Docker compose (alternative to steps 1–3)

Requires Docker with Compose.

docker compose up -d --build

This starts Ollama (localhost:11434) and the GLM-OCR layout service (localhost:5002) using the official ollama/ollama image. Models are pulled automatically:

• glm-ocr:latest — OCR (Z.ai GLM-OCR))
• kaelri/hy-mt2:1.8b — translation (Tencent Hy-MT2-1.8B)

Verify:

curl http://localhost:5002/health
curl http://localhost:11434/api/tags

The extension defaults (http://localhost:11434/, layout on port 5002) work without changes. The glm-ocr container reaches Ollama via Docker service DNS using GLMOCR_OLLAMA_ENDPOINT=http://ollama:11434 (set in docker-compose.yml). The first layout OCR request may download the Hugging Face layout model into the glm_ocr_models Docker volume.

---

1. Pull the OCR & translation models

ollama pull glm-ocr:latest
ollama pull kaelri/hy-mt2:1.8b

2. Start Ollama

Windows:

npm --prefix ./ollama run start:ollama

Linux/macOS:

./ollama/start-ollama.sh

3. Start the GLM-OCR service

Windows:

cd glm-ocr
.\run-service.ps1

Linux/macOS:

cd glm-ocr
./run-service.sh

Service health: http://localhost:5002/health

4. Build and load the extension

cd extension
npm install
npm run build:extension

Load the unpacked extension from the extension/ folder in Chrome. Set the Ollama endpoint in the options page (default: http://localhost:11434/).

Testing

See README.TESTS.md for the full test matrix.

Quick checks:

npm --prefix ./extension run test:unit
npm --prefix ./ollama run test:negative
pytest glm-ocr

Documentation

• docs/ENDPOINT_API_REFERENCE.md — HTTP API for Ollama and GLM-OCR endpoints
• extension/README.md — extension development
• glm-ocr/README.md — Python service setup
• ollama/README.md — Ollama dev tooling

License

MIT