HKUDS/DeepTutor

What it does

An 'agent-native' learning platform from HKU's HKUDS lab: chat tutoring, quizzes, deep research, visualization, problem solving and Mastery-Path practice all run on one agent loop. It supports multi-engine RAG knowledge bases (LlamaIndex, PageIndex, GraphRAG, LightRAG — even your Obsidian vault), lets you call Claude Code / Codex sub-agents from any conversation, and ships three-layer inspectable memory (L1 trace / L2 summary / L3 synthesis) with a Memory Graph for provenance.

What's inside

A full web app (Python backend + Next.js 16 frontend) with READMEs in 11 languages, Docker deployment, CLI/agent-native interfaces and the EduHub community skills ecosystem; releases ship very frequently.

Tech stack

Python 3.11+ + Next.js 16, multi-engine RAG + MCP tools + FAISS; Apache-2.0; created late December 2025 with an arXiv paper behind it.

Use cases

• Students & self-learners — feed in textbooks and papers, get a 24/7 personal tutor
• Exam candidates — prepare systematically with quizzes and the mastery path
• PKM enthusiasts — plug your Obsidian vault straight in for RAG Q&A

Why it's trending

It combines AI education with an agent-native workflow and carries the credibility of a university research lab, making the concept easy to evaluate and share.

Current chart placements