Repository navigation
MASHINA 1.5 — Local OCR
MASHINA 1.5.0
Fully local text OCR for receipts, workshop invoices/work orders, dashboard photographs and arbitrary screenshots.
- Read JPEG/PNG/WebP or a selected PDF page on the owner's server using bundled Tesseract rus+eng and Poppler. No external provider, paid API, cloud fallback or runtime model download.
- Inspect text, edit suggestions and explicitly select fields to apply to fuel, expense, maintenance or odometer drafts. All checkboxes start unchecked. Original attachment is opt-in.
- Recognition never saves a financial/mileage record. Normal Save preserves exact validation, historical mileage warnings and idempotency. Full tank, currency, trip/range/miles and ambiguous numbers are not guessed.
- Bounded on-demand processing, one job/CPU thread, private temporary files, time/resource limits, authorization and post-processing revocation checks.
- Updated roadmap includes general screenshots, workshop line-item extraction, crop/rotation and local-only future improvements.
Local OCR guide and limits · Roadmap
First local text-recognition slice, not universal document understanding. One PDF page per request; workshop line items remain manual from extracted text. Seven-segment/unlabelled/glare-heavy dashboards may require manual input. Offline OCR needs the local server; ordinary manual offline entry continues. No schema migration or infrastructure change. Real-engine Linux CI is mandatory before publication; Windows local native OCR is not installed.
Финальный Linux Docker CI успешно пройден: https://github.com/kosavan/mashina/actions/runs/38245792748
Проверенный исходный коммит: 2392dde998cda9bdb413164f44a8a2ce97a9d12a.