xby-recog

Pulling structured data out of real-world images—scanned paperwork, hand-written forms, national ID cards, passports, bank cards, license plates—remains one of the most repetitive and error-prone steps in production systems. Manual transcription is slow, and off-the-shelf OCR engines rarely return field-level accuracy on domain-specific documents. xby-recog plugs the xby recognition service directly into DeepSeek Harness, letting developers call professional image-recognition APIs without leaving the conversational workflow.

The plugin is organised around three capability families. The first is general text OCR, offered in a speed-oriented tier and a high-accuracy tier, both handling documents, signage, and screenshots. The second is handwriting-dedicated OCR with its own text-line detection, aimed at notes, signatures, and hand-filled forms. The third family covers structured document and ticket recognition: license plates (with colour detection and bounding boxes), national ID cards (automatic front/back discrimination and ID-number validation), passports and HK/MO/TW travel permits (including MRZ machine-readable zone parsing), bank cards (with Luhn check-digit verification), business licences, driving licences, and vehicle registration certificates. Every tool accepts three input formats—image URL, Base64-encoded string, or local file path—and a single API-key call persists across sessions and restarts.

Developers building automated document review, bulk data-extraction, or form-digitisation pipelines will find the one-tool-per-capability design especially convenient, because multi-step recognition workflows can be composed conversationally without writing custom API wrappers. It is a practical fit for teams in customer service, finance, logistics, or government who need reliable, field-accurate image recognition inside an existing Harness project.