xby-recog

Extracting structured data from images is a routine yet tedious task in production—scanned documents, handwritten forms, ID cards, passports, bank cards, license plates—where manual entry is slow and error-prone, and generic OCR cannot reliably parse field-level details on specific documents. xby-recog bridges that gap by bringing the xby recognition API into DeepSeek Harness as a plugin, so professional-grade image recognition can be invoked directly within a conversational workflow.

The plugin spans three categories of recognition. First, general text OCR in two tiers: a speed-optimised mode and a high-accuracy mode, both suited to documents, signage, and screenshots. Second, handwriting-specific OCR with dedicated text-line detection for notes, signatures, and hand-filled forms. Third, structured document and ticket recognition covering license plates (colour and bounding box), national ID cards (auto front/back detection and ID validation), passports and HK/MO/TW travel permits (with MRZ machine-readable zone parsing), bank cards (Luhn checksum validation), business licences, driving licences, and vehicle registration certificates. Every tool accepts three input formats—image URL, Base64 string, or local file path—and the API key persists automatically across restarts after a one-time setup.

It is well suited to developers building automated document review, data extraction, or form-digitisation pipelines, as well as application engineers handling bulk image-to-text workloads in customer service, finance, logistics, or government workflows. Because each capability maps to a single named tool, teams can compose multi-step recognition workflows without writing bespoke API wrappers, keeping the whole pipeline inside the Harness conversation loop.