问道：面向学术策展的 AI 辅助套件
问道不是论文代写器，而是帮助研究者在 AI 时代保持主体参与的学习、提问、求证与策展发布工具。它服务于从论文视图到完整学术作品的转换：公共导读、数据说明、历史版本、媒体材料和机器可读格式会在同一工作流中被组织起来。
对话式研究工作台
与所选 AI 模型对话完成文献、大纲、成稿与配图，来回修改的每一版自动成为历史版本。立即体验预览版。
阅读助手
把艰深论文和著作转化为概念图、问题树和争论谱系，同时保留值得细读的原文入口。
选题对话
通过启发式追问帮助研究者辨认真正关心的问题、异常和长期风险。
策展发布包
同步生成论文视图、公共导读、数据说明、媒体材料、版本说明和 AI 可读格式。
路线图
问道以成熟的开源组件组合而成，从预览逐步走向可用。以下方向按优先级推进。
真实模型接入
通过开源多模型网关（one-api / new-api 或 LiteLLM）统一接入 Claude、OpenAI、DeepSeek、GLM、通义千问等，向导对话由真实模型驱动，脚本模式保留为离线回退。
会员与自备密钥
注册会员可使用平台提供的模型额度，也可以填入自己的 API 密钥直连所选模型；密钥只保存在本地浏览器，不经平台存储。
直通发布
向导产出的发布包直接提交到平台，以待审核状态进入编辑流程；管理员批准后纳入某期期刊，或由策展重新组合展示后公开。
本地版
与 Cowork、Codex 等本地智能体配合，调用本地文献、数据与计算资源完成策展，与在线版共用同一发布包格式。
Wendao: AI assistance for scholarly curation
Wendao is not a paper ghostwriter. It helps researchers stay actively involved while reading, questioning, verifying, and curating for publication. It supports the move from a paper view to a complete scholarly work: public guides, data notes, version records, media assets, and machine-readable formats in one workflow.
Curation Wizard · Online Preview
Choose an AI model, then move through one continuous dialogue — idea, literature, approach, outline, draft, figures — with version history precipitating automatically and a platform-ready package at the end.
Dialogue research workbench
Work with your chosen model through literature, outline, draft, and figures — every revision round becomes a version automatically. Try the preview.
Reading assistant
Turn difficult papers and books into concept maps, question trees, and debate maps while preserving key passages for close reading.
Question dialogue
Use heuristic dialogue to identify the questions, anomalies, and long-term risks that a researcher truly cares about.
Curation package
Generate paper views, public guides, data notes, media assets, version notes, and AI-readable formats together.
Roadmap
Wendao is assembled from proven open-source components and moves from preview to production in stages.
Real model access
An open-source multi-provider gateway (one-api / new-api or LiteLLM) will connect Claude, OpenAI, DeepSeek, GLM, Qwen and more, driving the dialogue with real models; the scripted mode remains as an offline fallback.
Members & bring-your-own key
Registered members can use platform model quotas or enter their own API keys to connect directly; keys stay in the local browser and are never stored by the platform.
Direct publishing
The wizard’s package submits straight to the platform as a pending work; once approved it joins a journal issue or a curated exhibition and goes public.
Local edition
Local agents such as Cowork or Codex can bring local literature, data, and compute — sharing the same package format as the online edition.