wendao 0.3.1

Wendao (问道, wèn dào, "asking the way") is an AI learning companion. It turns your course materials into two things for students:

• an interactive knowledge graph of the course's chapters and concepts, and
• an AI learning agent that answers questions using only the course materials, with a link to the page each answer came from.

It was built for the NUS course MLE4217/5219 Materials Informatics, and this repository uses that course as its example. Nothing in the code is tied to that course, so you can run it on any course written as Markdown pages and Jupyter notebooks.

Guides: teacher guide · server setup · developer notes · plans and ideas

Demo

Knowledge graph

Chapters, pages, and concepts, linked where the course connects them. Drag, zoom, and search; click a concept to read a short introduction from the notes and jump to the course page.

_{Full video: knowledge_graph_demo.mp4}

AI learning agent

Select a node and press Explain, or ask your own question. Answers come only from the course and link to the pages they used. If the course doesn't cover a question, the agent says so.

_{Full video: ai_learning_agent_demo.mp4}

Chat widget on your course website

Install

You need Python 3.11 or newer.

pip install wendao

Students don't install anything: they use your website, the chat widget on your course pages, or one HTML file.

Quick start

wendao init my-course        # a new workspace with a notes/ folder
# copy your notes (Markdown, notebooks, PDF, slides, Word, LaTeX) into my-course/notes/
cd my-course
# choose a model under [model] in wendao.toml, and put its API key in .env
wendao build                 # graph, concepts, and search index
wendao doctor                # checks everything and says how to fix problems
wendao serve                 # opens the course website at http://127.0.0.1:5057/

No model yet? Everything except the AI answers still works, and wendao ask --search-only "..." shows what search finds. To try the example course: git clone https://github.com/deng-group/wendao.git, then cd wendao/examples/mle4217_5219 and run wendao serve --no-check (it is already built; for AI answers, set your own model under [model] in its wendao.toml).

The teacher guide explains each step and setting.

Share it with students

You decide who pays for AI answers: your key (with a daily limit per student and optional sign-in), each student's own key, or either. See How students use AI.

Commands

Run any command with --help to see its options.

How it works

Wendao splits your notes into small pieces, links chapters and concepts that appear together, and, for each question, finds the most relevant pieces with keyword and meaning-based search. If they don't really answer the question, it says so instead of guessing. Otherwise a language model (Claude, GPT, Gemini, DeepSeek, or a local model) writes the answer from them. No training or GPU is needed.

Citation

Wendao is developed by the Deng group in the Department of Materials Science and Engineering at the National University of Singapore, for MLE4217/5219 Materials Informatics. If you use Wendao or adapt it for your own course, please cite this repository:

@software{wendao2026,
  author = {Deng, Yanhao and Deng, Zeyu},
  title  = {Wendao: an AI learning companion with an interactive course knowledge graph},
  year   = {2026},
  url    = {https://github.com/deng-group/wendao}
}

License

Wendao is free software under the GNU General Public License v3.0 or later. You can use, study, change, and share it. If you share a changed version, you must share its source code under the same license.