DeepTutor: The Open-Source AI Tutoring Platform That Grows With You — 10,000+ GitHub Stars

What Is DeepTutor and Why It Matters

The education landscape is shifting. Static courses and one-size-fits-all tutorials can't keep up with how developers actually learn today. Enter DeepTutor — an open-source, agent-native platform that reimagines personalized tutoring for the AI era.

Built by the Hong Kong University Data Science Lab (HKUDS), DeepTutor isn't just another chatbot wrapper. It's a comprehensive learning ecosystem with eight integrated surfaces, three-layer memory architecture, and support for multiple RAG engines. The project has gained significant traction on GitHub with over 10,000+ stars and active daily development.

For developers learning to code, DeepTutor represents something powerful: a self-hosted, privacy-respecting alternative to proprietary AI tutoring platforms — one you can customize, extend, and own completely.

Eight Surfaces, One Unified Learning Experience

Most AI learning tools give you a chat window and call it a day. DeepTutor takes a fundamentally different approach with eight distinct but interconnected surfaces:

1. Home (Chat Workspace)

The default agent loop where you interact with your tutor. It supports tools, retrieval from your knowledge base, file attachments, and persistent memory — all in one session. Think of it as your primary learning conversation.

2. Partners

Persistent AI companions with their own personality ("soul"), knowledge library, and IM channel integrations (Discord, Telegram, Slack, Mattermost, and more). Each partner maintains private memory and can be assigned to specific users in multi-tenant deployments.

3. My Agents

Connect your local Claude Code or Codex agents and import past conversations to consult them live. This bridges your existing coding assistants with DeepTutor's learning infrastructure.

4. Co-Writer

A selection-aware Markdown workspace for collaborative editing grounded in your knowledge base or web sources. Perfect for writing documentation, study notes, or technical articles with AI assistance.

5. Book (Interactive Books)

Compile knowledge bases and notebooks into interactive "living books" with embedded quizzes, references, and chat capabilities. Turn your study materials into engaging, interactive learning experiences.

6. Learning Space

A reusable hub for skills, personas, notebooks, and question banks. Organize your learning resources and track progress across multiple topics or courses.

7. Memory

Inspectable, three-layer memory (L1/L2/L3) that keeps personalization transparent and auditable. Unlike black-box systems, you can see exactly what DeepTutor remembers and why.

8. Knowledge Center

Versioned, multi-engine retrieval libraries supporting GraphRAG, PageIndex, LightRAG, and FAISS vector backends. Upload PDFs, DOCX, XLSX, PPTX files and build sophisticated retrieval-augmented generation systems.

Technical Architecture: Built for Developers

DeepTutor's architecture reflects its open-source, developer-first philosophy:

Flexible RAG Engines

Choose from multiple retrieval engines based on your needs:

• LlamaIndex — Production-grade RAG with multimodal image extraction
• LightRAG — Lightweight indexing without heavy dependencies
• GraphRAG — Knowledge graph-based retrieval for complex relationships
• FAISS — Fast vector similarity search for large knowledge bases

Three-Layer Memory System

DeepTutor implements a sophisticated memory architecture:

• L1 (Working Memory) — Current conversation context
• L2 (Session Memory) — Recent interactions and preferences
• L3 (Long-term Memory) — Persistent knowledge and learning patterns

Multi-Provider Support

Works with any OpenAI-compatible API, plus native support for:

• OpenAI (GPT-4, GPT-3.5)
• Anthropic (Claude)
• Google (Gemini)
• NVIDIA NIM
• Local models via Ollama, LM Studio, or Lemonade

Real-World Example: Building a Python Learning Companion

Let's say you're learning Python and want a personalized tutor. Here's how you'd set up DeepTutor:

# 1. Install DeepTutor
mkdir python-tutor && cd python-tutor
pip install -U deeptutor
deeptutor init

# 2. Configure your preferred LLM (e.g., Claude)
# Edit .env or use the web UI at http://localhost:3782

# 3. Start the platform
deeptutor start

Now upload your Python learning materials:

• Upload "Python Crash Course" PDF to the Knowledge Center
• Add your coding exercises as Markdown files
• Create a Partner named "PyMentor" with a friendly, encouraging personality

Start chatting with PyMentor:

You: "I'm struggling with list comprehensions. Can you explain them with examples?"

PyMentor: [Retrieves relevant sections from your uploaded book, provides explanations 
with code examples, and creates practice problems based on your skill level]

You: "Show me a real-world use case"

PyMentor: [Uses its memory of your interests to provide relevant examples, 
then adds the interaction to long-term memory for future reference]

Over time, PyMentor learns your learning style, remembers which concepts you've mastered, and adapts its teaching approach accordingly.

Key Benefits for Developers

• Fully Open Source — MIT licensed, no vendor lock-in, complete control over your data
• Self-Hosted — Run on your laptop, home server, or cloud infrastructure
• Privacy-First — Your learning data never leaves your infrastructure
• Extensible — Plugin system for custom tools, skills, and integrations
• Multi-User Ready — Deploy for teams or classrooms with isolated workspaces
• Active Development — Frequent releases with new features (v1.5.1 released July 9, 2026)
• Rich Ecosystem — IM integrations, document parsing, visualization tools, and more

Getting Started in 5 Minutes

DeepTutor offers four installation paths. Here's the fastest:

# Quick install (Python 3.11+)
mkdir my-tutor && cd my-tutor
pip install -U deeptutor
deeptutor init
deeptutor start

# Open http://localhost:3782 in your browser

For Docker users:

docker run --rm --name deeptutor \
  -p 127.0.0.1:3782:3782 \
  -v deeptutor-data:/app/data \
  ghcr.io/hkuds/deeptutor:latest

After installation:

1. Configure your LLM provider in Settings
2. Upload learning materials to the Knowledge Center
3. Create a Partner with your preferred teaching style
4. Start learning!

Why DeepTutor Stands Out

Compared to other AI tutoring solutions:

• vs. ChatGPT/Claude — DeepTutor offers persistent memory, structured learning spaces, and RAG over your materials
• vs. Khan Academy/Socratic — Self-hosted, customizable, and works with any LLM provider
• vs. Building your own RAG — Production-ready with polished UI, no need to reinvent the wheel

Frequently Asked Questions

Is DeepTutor really free?

Yes, DeepTutor is completely open source under the MIT license. You only pay for the LLM API usage (or use local models for free).

Can I use it offline?

Yes! DeepTutor supports local models through Ollama, LM Studio, or any OpenAI-compatible local server. Once installed, it works without internet.

Does it support multiple languages?

Yes, DeepTutor works with any language your chosen LLM supports. The UI is available in English and Chinese, with community translations in progress.

Can multiple people use the same instance?

Absolutely. DeepTutor supports multi-user deployments with isolated workspaces, admin controls, and per-user resource isolation.

What document formats does it support?

PDF, DOCX, XLSX, PPTX, Markdown, plain text, and images (with multimodal extraction). The platform handles parsing and indexing automatically.

How does the memory system work?

DeepTutor uses a three-layer memory architecture (L1/L2/L3) that's fully inspectable. You can view, edit, or delete any memory entry through the Memory interface.

Can I integrate it with Discord or Telegram?

Yes! Partners support 15+ IM channels including Discord, Telegram, Slack, Mattermost, Matrix, and more.

What's the difference between Partners and My Agents?

Partners are persistent companions with their own personality and knowledge base. My Agents lets you connect external coding assistants (Claude Code, Codex) and import their conversation history.