Versions:
AutoDocxProofreading, published by CZ600 and currently available at version 1.2.1 across two released versions, is a desktop application in the document proofreading and text editing software category that uses large language models to proofread Word documents, adjust formatting, and reduce AI detection rates in text. Designed primarily for proofreading long documents and optimizing academic paper formatting, the software detects typos, punctuation errors, grammar problems, and text consistency issues within Word files, offering revision suggestions to the user. To address the forgetting and hallucination problems that large models exhibit when handling lengthy documents, the application implements a purpose-built architecture intended to improve proofreading accuracy, along with a parallel processing architecture that significantly increases processing speed for long documents, and it allows the corrected document to be exported with a single click. The software also incorporates a local knowledge base with retrieval-augmented generation (RAG) support, providing reference material for the model during proofreading. Its AI-rate reduction feature operates in a segmented parallel manner, dividing a document by natural paragraphs and invoking the large model to process them concurrently, while its text polishing capabilities further help lower AI detection scores. Additionally, the application supports cross-document format cloning, enabling one-click adjustment of formatting throughout an entire document without requiring the large model, which makes formatting operations faster. Compared with using Claude Code skills for proofreading and formatting, the software offers a visual interface for directly inspecting results with greater speed and lower token consumption; compared with web-based AI applications such as ChatGPT or Doubao, it provides a one-click workflow that eliminates repetitive back-and-forth interaction; and compared with the built-in proofreading systems of Microsoft Office and WPS, it delivers more intelligent checking with a broader range of recognized error types.
Tags: