Document OCR System

API models: 0 RAM | Chhagan local: loads only when selected

Select Model

API (0 RAM):

  • Qwen3-VL-8B — 32+ languages
  • Kimi-K2.5 — fastest

Local (~8GB, lazy):

  • Chhagan ML-VL-OCR — specialized
256 4096

Features: MRZ auto-parse | Face detection | Multi-page PDF | 32+ languages | GPU/CPU/MPS auto