User guide

This guide covers StyleOCR features beyond the basics in Getting started.

Recognition modes

• Fast mode — tuned for speed on clear scans and simple pages.
• Pro mode — higher accuracy when you can spend more time per page.
• VLM mode — uses a vision-language model for flexible recognition when layout, tables, or mixed content need deeper understanding.

You can switch modes per task. If results look noisy, try Pro or VLM before changing models.

Vision models (OCR vs VLM)

Under Settings → Vision models:

• OCR models extract plain text from images and scans.
• VLM models interpret layout, tables, and structure for richer exports.

Models are downloaded on demand and stored on your machine. You can install multiple bundles, switch the active model, or delete downloads to free space. Model cards may indicate capabilities such as tables, formulas, seals, or handwriting—enable what matches your documents.

Document output

Common export targets include Word (.docx), Markdown, plain text, and Excel (.xlsx).

Useful options (where available in your build):

• Style extraction — preserve more visual structure in the exported document.
• Retain images in document — keep page images alongside recognized text.
• Page break mode — continuous flow, one page per image, or match PDF pagination.

Primary language settings help the engine pick the right script or language family for multilingual stacks.

Information extraction

Beyond full-page OCR, StyleOCR can run structured extraction (for example invoices, passports, ID cards) using guided fields.

• Use built-in presets when they match your form type, or add custom modes with clear field names (specific labels work better than generic words like “number”).
• Custom modes support a name, description, and a list of fields the model should fill.

Extraction runs locally with the same privacy posture as normal OCR.

Tasks, history, and background work

• Each task tracks files, progress, and results. You can search past tasks from history.
• Long jobs can run in the background; notifications can alert you when a task completes (see Settings → Notifications).
• Depending on version, you may stop a running task and resume later from history.

Agent sidebar

The Agent panel drives VLM-oriented workflows such as recognize, translate, extract, and summarize. Some flows require a short prompt (for example target language for translation or field instructions for extraction). Agent output streams into the task UI so you can review before exporting.

MCP server

Settings → MCP server lets you run a local MCP endpoint so compatible AI assistants connect to StyleOCR on localhost. This does not upload your documents to a third-party SaaS by itself—it exposes an interface on your machine for tools you explicitly configure.

Restart the service when you change the listening port. Treat the displayed URL as sensitive on shared computers.

Appearance and language

Settings → Appearance controls theme (system / light / dark) and accent color. Language & region changes the UI locale.

Billing and plans

Subscription or trial flows (if enabled in your build) live under Settings → Billing & Plans. Purchases and restores follow the rules of the platform store you used. Free tiers may apply daily quotas; the app surfaces upgrade options when limits are hit.

Feedback and logs

Settings → Feedback opens a mail draft to the developer. Optionally attach app logs to speed up diagnosis.

---

For how data stays on device and when network access occurs, see Security architecture.