{% extends "base.html" %} {% block content %}

Settings

Config: {{ config_path }}
{% if status_level == 'full_local' %}
๐ŸŸข FULLY LOCAL โ€” All processing on this machine. Nothing leaves.
{% else %}
๐ŸŸก PARTIALLY LOCAL โ€” Data stored here, but sent to {{ cloud_model }} for extraction.
{% if needs_privacy_ack %}
Cloud mode is selected but not acknowledged. Upload/extraction will be blocked until you confirm below.
{% endif %} {% endif %}
{% if error %}
{{ error }}
{% endif %}
Model backend
Local model (Ollama)
{% if not selected_is_vision and cfg.model_backend == 'local' %}
โš ๏ธ {{ cfg.local_model }} is a text-only model โ€” it cannot read document images and will return empty extractions. Switch to a vision model (e.g. glm-ocr โ€” run ollama pull glm-ocr) or use Cloud mode.
{% endif %}
Vision models (โœ…) can read document images. Text-only models (โš ๏ธ) will not extract any data.
Cloud model
๐Ÿ“Š How model choice affects extraction quality
Local (Ollama)
โœ… Free ยท โœ… Fully private ยท your data never leaves this machine
โš ๏ธ Smaller models may struggle with:
  • Complex multi-copy forms (consolidated 1099s, K-1s)
  • Low-quality scans or non-standard layouts
  • Handwritten or corrected fields
Typical confidence: 0.85โ€“0.97 with glm-ocr on clean PDFs
Cloud (Claude / OpenAI)
  • โœ… Higher accuracy on complex forms
  • โœ… Better handling of edge cases and poor scans
  • โœ… Confidence typically 0.85โ€“0.97
โš ๏ธ Your document content leaves this machine โ€” see Privacy Policy
โš ๏ธ Requires API key or cloud subscription
๐Ÿ’ก Tip: glm-ocr (ollama pull glm-ocr) is the recommended local model โ€” purpose-built for document OCR, outperforms llava significantly. Switch to cloud only for low-confidence results or complex multi-page K-1s.
{% endblock %}