ArtificeOCR turns a folder of scanned or photographed pages into corrected, searchable text — and, where the source language isn't English, a translation — without sending a single page to a remote service. Add files by drag-and-drop, by browsing to a folder, or by pulling items directly from a read-only Tropy project; queue them; choose which stages to run; and watch the queue update live as each document moves through the pipeline.

Pipeline stages

OCR — reads each page image with a vision-capable model. The default route is a local model served by LM Studio; Ollama, Hugging Face and a cloud API key are configurable alternatives in Settings, one per stage.
Cleanup — a language model pass corrects recognition errors and normalises layout while a content-preservation guard keeps the original wording whenever the edit looks lossy, so a page is either cleaned or left untouched, never silently reworded.
Translate — renders the cleaned text into English. A confident language-detection pass skips already-English pages rather than risking an LLM "helpfully" rewriting text that didn't need translating.
Export — compile any stage's output into a continuous reading PDF or Markdown file, send corrected text back to Tropy as notes or transcriptions, or export a collection for LudwigLang's Import page.

Preview, history and correction

The Preview tab shows the source scan beside raw OCR, cleaned and translated text side by side, with the original always one click away behind "Original" so a correction never hides what the model actually produced. History keeps every past run and document searchable, with the same compare view and a thumbnail strip across the source images. Batch Correct applies a single find/replace across the whole queue or a selection, with its own template list for corrections you make often.

Templates and quick actions

Settings → Run Templates saves a stage-and-option combination under a name, so a recurring job configuration is one click instead of six. Ctrl+K (Cmd+K on macOS) opens a command palette that jumps to any tab, runs a queue action, or applies a template without leaving the keyboard. The same palette also toggles light and dark themes, as does the icon at the right of the top navigation bar.

Design language

This interface implements The New Masses design system: an offline-first, print-inflected editorial aesthetic with warm paper surfaces, a deep teal accent, and fluid serif typography. All design tokens are served from the shared packages/shared-ui/shared_ui/assets/tokens.css; never hardcode colours or spacing.

Run it

From the project root:

pip install -e ".[web]"
artifice-ocr-web
# → http://localhost:8765

ArtificeOCR · an Artifice platform · FastAPI + Jinja2 · vanilla JS.