DeepTutor: HKU's Agent-Native Tutoring Workspace With Inspectable Memory

Vibe-Trading: HKU’s Open-Source Research Agent for Markets, Not a Trading Bot • nanobot: HKU’s Ultra-Lightweight Self-Hosted Personal AI Agent

What Is DeepTutor? #

DeepTutor is the third HKUDS project covered on dibi8, alongside Vibe-Trading and nanobot — same Hong Kong University Data Science lab, this time applied to education. It’s an open-source (Apache-2.0) agent-native learning workspace, backed by a published arXiv paper, that connects tutoring, problem solving, quiz generation, research, visualization, and mastery practice into one system with shared memory and context.

🔗 GitHub: https://github.com/HKUDS/DeepTutor 🌐 Homepage: https://deeptutor.info

First pushed in December 2025, DeepTutor reached 30,800+ GitHub stars by late July 2026, with an active release cadence — v1.5.6 shipped the day before this article, per the project’s own changelog.

Key Features #

• One runtime for every mode — Chat, Quiz, Research, Visualize, Solve, and Mastery Path all run on the same agent loop, so switching what you’re doing doesn’t lose context on who the learner is
• Connected learning context — knowledge bases, books, Co-Writer drafts, notebooks, question banks, personas, and memory stay shared across every workflow instead of living in separate tools
• Subagents and Partners — consult a live coding CLI (Claude Code, Codex, Gemini, Kimi, opencode, or MiMo) from any turn, or run persistent IM companions (“Partners”) on the same underlying memory
• Multi-engine knowledge — versioned RAG libraries across LlamaIndex, PageIndex, GraphRAG, or LightRAG, plus a linked Obsidian vault option
• Extensible tools and skills — built-in tools, MCP servers, image/video/voice generation, and installable community skills via EduHub
• Inspectable memory — L1 traces, L2 surface summaries, and L3 synthesis layers make personalization visible and editable, with a Memory Graph tracing every claim back to its source evidence

Install #

Fastest path — PyPI (requires Python 3.11-3.13, Node.js 20+):

mkdir -p my-deeptutor && cd my-deeptutor
pip install -U deeptutor
deeptutor init     # prompts for ports + LLM provider + optional embedding
deeptutor start    # starts backend + frontend

Open the printed frontend URL — by default http://127.0.0.1:3782. Ctrl+C stops both processes. Skipping deeptutor init is fine for a quick trial; configure providers later under Settings → Models.

From source (Python 3.11-3.13, Node.js 22 LTS to match CI/Docker):

git clone https://github.com/HKUDS/DeepTutor.git
cd DeepTutor
python3 -m venv .venv && source .venv/bin/activate
python -m pip install --upgrade pip
# install backend + frontend deps per docs

The CLI Is Built to Be Driven by Other Agents #

One deeptutor binary, two interfaces: an interactive REPL for humans, and structured JSON for agents driving it as a tool — same capabilities, tools, and knowledge bases either way.

Interactive:

deeptutor chat                                              # interactive REPL
deeptutor run chat "Explain the Fourier transform" --tool rag --kb textbook

Agent-driven (NDJSON output):

deeptutor run deep_solve "Find d/dx[sin(x^2)]" --tool reason --format json

Add --format json to any run command and DeepTutor streams NDJSON — one event per line (content, tool_call, tool_result, done), each tagged with its session_id. Runs are headless-safe: an ask_user pause with no TTY auto-resolves with an empty reply instead of hanging forever.

Recent Release Highlights (Per the Project’s Own Changelog) #

That “non-English languages no longer collapse to Chinese” fix in v1.5.6 is a real, specific bug the project shipped a fix for — worth knowing if you hit odd language behavior on an older version.

Use Cases #

1. Studying a Subject With Memory That Persists Across Sessions #

Ask questions over weeks or months and have DeepTutor’s L1/L2/L3 memory layers keep track of what you’ve already covered, without starting from zero each session.

2. Building Course Materials With Co-Writer + Knowledge Bases #

Combine a versioned RAG knowledge base with the Co-Writer drafting tool to produce material grounded in specific source documents, with traceable citations.

3. Letting an Agent Drive Tutoring as a Backend Service #

Use deeptutor run ... --format json to integrate DeepTutor’s tutoring/research capabilities into a larger agent pipeline rather than using the web UI directly.

4. Consulting a Coding CLI Mid-Lesson #

Pull in Claude Code or another coding CLI as a subagent from within a learning session when a topic turns into “show me working code.”

Related Repositories #

Related Articles #

• Vibe-Trading: HKU’s Open-Source Research Agent for Markets, Not a Trading Bot — same lab, a completely different domain
• nanobot: HKU’s Ultra-Lightweight Self-Hosted Personal AI Agent — same lab’s take on a general personal agent

Conclusion #

DeepTutor treats personalized tutoring as a memory-and-context engineering problem rather than a prompt-engineering one — one agent loop across every learning mode, RAG grounded in your own materials, and a memory system designed to be inspected rather than trusted blindly. Backed by a published paper and an unusually active release cadence (four releases in the week before this article), it’s a serious academic-lab project, not a weekend hack.

Best for: Learners who want long-term, context-persistent tutoring across a subject, and developers interested in integrating an agent-drivable tutoring/research backend (via the JSON CLI) into a larger system.

GitHub: https://github.com/HKUDS/DeepTutor

Recommended Infrastructure for Self-Hosting #

If you want DeepTutor’s backend and frontend running persistently instead of on a personal laptop:

• DigitalOcean — $200 free credit for 60 days across 14+ global regions.
• HTStack — Hong Kong VPS with low-latency access from mainland China. This is the same IDC that hosts dibi8.com.

Affiliate links — they don’t cost you extra and they help keep dibi8.com running.

Last updated: 2026-07-29

References & Sources #

• DeepTutor
• DeepTutor homepage
• DeepTutor arXiv paper
• Vibe-Trading
• nanobot