Human-AI Co-Writing With Full Accept/Reject Control on Every Edit

Most AI writing tools work like a black box. You highlight text, hit a button, and the AI rewrites your paragraph. Maybe it improved things. Maybe it nuked a sentence you liked. You can undo, but you lost visibility into what changed and why.

We built CoCraft to fix that. Part of the OCTO ecosystem, Apache 2.0 licensed. The core idea: AI is your writing partner, not your ghostwriter.

How the edit flow works

Select text. Tell the AI what you want in plain language, like 'make the claim less absolute' or 'add a disclosure about AI usage per journal policy.' AI generates an edit proposal, CoCraft shows a diff. Every change is highlighted. You accept or reject each one separately.

This matters more than it sounds. Revising a paper after peer review, you might get three comments touching the same section. One asks for an AI usage acknowledgment. Another flags your 'near-optimal' claim lacks empirical backing. A third points out a concurrent result computing the theoretical upper bound that you did not reference.

CoCraft lets you handle each comment, generate edit proposals, and go through diffs one change at a time. Keep what works, toss what does not.

Compilation error diagnosis

CoCraft has a diagnostic panel for compilation errors. Panel locates the error, AI proposes a fix in one step. Small feature, saves real time.

Connected to Octo

CoCraft connects into the Octo ecosystem, so collaborator discussions retain full context. Your co-author's notes about why a paragraph was phrased a certain way stay visible when the AI proposes changes.

Trade-offs we are honest about

The diff-based flow adds friction compared to just rewriting it all. For quick drafts where you do not care about granular control, it can feel slower. We think the trade-off is worth it for serious writing like papers and technical docs, but it is not the right tool for every use case.

Source code and ecosystem: OCTO on GitHub | Mano-P

相关推荐
泯泷1 小时前
AI 怎样"查资料"?——RAG 检索增强生成入门(三)
人工智能·llm
Solange__2 小时前
Blender 3D模型优化:普通复制、共享网格、实例、合并怎么选?
人工智能·3d·ai·自然语言处理·编辑器·blender·材质
师夷长技以自愈2 小时前
手搓 AI Agent 第 8 篇:从手写循环到画流程图——LangGraph 封装
人工智能·python
数据治理自习室2 小时前
2026最新:MCP开发与使用指南
人工智能
小小疯狐狸2 小时前
我踩过的Agent harness几个坑:上下文越跑越慢
人工智能
kaixin_啊啊2 小时前
【零基础学AI】第 30 章——对话记录与结构化输出
人工智能
泯泷3 小时前
让知识"长骨架"——RAPTOR、GraphRAG 与 OpenViking(四)
人工智能·llm
kaixin_啊啊3 小时前
【零基础学AI】第 30 章课后练习与答案
人工智能
电科_银尘3 小时前
【深度学习项目实战·33】自定义数据集:把自己的图片接入PyTorch
人工智能·pytorch·深度学习
武子康4 小时前
单卡 A6000 跑 Qwen-Drive:怎样留下可比较的第一条基线?
linux·人工智能·深度学习