dsh-tool-ocr v0.1.0

dsh-tool-ocr v0.1.0

Standalone DeepSeek Harness plugin: local image text recognition for models without vision input (e.g. DeepSeek chat models), backed by the standalone nbocr (newbee-ocr) engine over PP-OCRv6 models.

Published on npm as dsh-tool-ocr.

Features

• ocr tool with recognize / status / install (alias of check) / check actions
• Image inputs: local path (resolved against the session workspace) or attachment_id
• Reading-ordered text, per-block bounding boxes, review flags (low-confidence / amount / numeric / date / quantity), heuristic Markdown table reconstruction
• Robust against MNN diagnostics polluting engine stdout; honors caller cancellation
• Fully out-of-tree: depends only on published dsh base packages

Install

Build the nbocr engine, then install the plugin into your dsh profile:

cargo build --release        # build nbocr from the newbee-ocr-cli repo
dsh plugin --profile web add dsh-tool-ocr
# or: cd ~/.dsh/profiles/web && pnpm add dsh-tool-ocr

Add the plugin row to your profile's cordis.patch.yml:

- insert:
    - id: ocr
      name: 'dsh-tool-ocr'
      inject: [tools, subprocess, systemPrompt]
      config:
        command: 'C:/path/to/nbocr.exe'
        detModel: 'v6-tiny'
        language: 'chinese'

See the README (English | 中文) for full configuration and usage.