HKUDS/DeepTutor
DeepTutor: Lifelong Personalized Tutoring. https://deeptutor.info/.
What it solves
DeepTutor provides a unified, agent-native learning workspace that eliminates the fragmentation between different educational tools. It integrates tutoring, problem solving, quiz generation, research, and visualization into a single system where the learner's context and memory are preserved across all activities.
How it works
DeepTutor operates on a single agent loop that powers multiple modes (Chat, Quiz, Research, Visualize, Solve, and Mastery Path). It utilizes a multi-engine knowledge system for RAG (Retrieval-Augmented Generation) using frameworks like LlamaIndex, GraphRAG, and LightRAG, and connects to various LLM providers. It also features a three-layer memory system (L1 traces, L2 surface summaries, and L3 synthesis) to maintain a lifelong personalized learning history.
Who it’s for
It is designed for learners and educators who want a personalized, intelligent tutoring experience that connects various study workflows—such as drafting with a Co-Writer, managing a question bank, and interacting with specialized "Partner" agents—within one environment.
Highlights
- Unified Agent Runtime: Switches objectives (e.g., from chatting to solving a problem) without changing the underlying engine.
- Multi-Engine RAG: Supports diverse knowledge retrieval methods including GraphRAG, LightRAG, and linked Obsidian vaults.
- Subagents and Partners: Ability to consult live coding CLIs (like Claude Code or Codex) or persistent IM companions during a session.
- Extensible Ecosystem: Supports MCP servers, community-installed skills from EduHub, and various multimodal generation models.
- Lifelong Memory: An inspectable three-layer memory architecture for deep personalization.