AI tutoring tools have been everywhere lately. Most of them are just ChatGPT wrappers with a different skin — ask a question, get an answer, move on. Nothing sticks. Nothing adapts. Nothing remembers.
DeepTutor, built by the HKUDS team at Hong Kong University, takes a fundamentally different approach. It's an agent-native, open-source learning platform that builds a persistent profile of how you learn, what you've studied, and where you're heading — then uses that context across every interaction.
It hit 10,000 GitHub stars in 39 days. The v1.0.0 release just dropped with a ground-up architecture rewrite (~200k lines). This isn't a weekend side project. It's infrastructure.
Why DeepTutor Is Different from Every Other AI Tutor
Most "AI education" products fall into one of two traps:
- Stateless Q&A bots — They answer your question and forget you exist. No continuity, no growth tracking
- Rigid curriculum systems — Pre-built paths that don't adapt. You're forced into someone else's learning sequence
DeepTutor avoids both by being agent-native from the ground up. Every feature — chat, quizzes, research, writing — shares a unified context. Your learning profile evolves with every interaction.
The Core Architecture
DeepTutor is built around a two-layer plugin model that separates Tools (atomic actions like web search, file parsing, math rendering) from Capabilities (higher-level skills like Guided Learning, Quiz Generation, Deep Research).
This isn't just clean architecture for architecture's sake. It means:
- New capabilities plug in without breaking existing ones
- Each TutorBot can have its own set of skills and personality
- The system can be extended by the community without touching core code
The stack is Python + Next.js 16 + React 19, with WebSocket-based real-time communication. It supports multiple LLM providers (OpenAI, Anthropic, Azure, and more) so you're not locked into a single vendor.
The Five Modes (And Why They Matter)
DeepTutor's unified workspace gives you five modes, all sharing the same conversation thread:
1. Chat
Standard conversational AI, but with access to your knowledge base and persistent memory. It knows what you've been working on.
2. Deep Solve
Multi-agent problem solving. When a question is too complex for a single response, DeepTutor escalates to a multi-step reasoning pipeline. Think of it as the "deep research" mode for tough problems.
3. Quiz Generation
Automatically generates quizzes from your uploaded materials. The v1.0.1 release added duplicate prevention with generation history — a small but critical detail that shows the team actually uses their own product.
4. Deep Research
Autonomous research mode that searches the web, synthesizes findings, and produces structured output. Great for literature reviews or exploring new topics.
5. Math Animator
Interactive mathematical visualizations. Particularly useful for STEM learners who need to see concepts, not just read about them.
TutorBots: Autonomous Tutors, Not Chatbots
This is the most interesting part. DeepTutor lets you create personal TutorBots — autonomous tutor agents that:
- Live in their own workspace with dedicated memory
- Have configurable personality and skill sets
- Can learn new abilities over time
- Set reminders and track your progress
- Powered by nanobot, an underlying agent framework
Each TutorBot evolves as you grow. It's not a static prompt — it's a learning companion that accumulates context about your strengths, weaknesses, and goals.
Knowledge Hub and RAG Pipeline
Upload PDFs, Markdown, or plain text files. DeepTutor builds RAG-ready knowledge bases that power every conversation.
What's notable:
- Flexible RAG pipeline — choose your embedding model and retrieval strategy per knowledge base
- Color-coded notebooks — organize insights across sessions
- Documents actively power conversations — they're not just stored, they're used
- MinerU integration for advanced document parsing (including nested outputs in v1.0.0-beta.2)
The embedding system handles HTTP 429 rate limits with automatic retry (added in v1.0.0-beta.4). Again, small detail that matters in production.
Self-Hosted, No Vendor Lock-In
DeepTutor runs entirely on your machine. The setup process is straightforward:
git clone https://github.com/HKUDS/DeepTutor.git
cd DeepTutor
conda create -n deeptutor python=3.11 && conda activate deeptutor
python scripts/start_tour.py
The start_tour.py script walks you through dependency installation, provider configuration (LLM, embedding, search), and live connection testing. It handles both web and CLI modes.
For Docker users, pre-built images are available on GHCR. Minimum requirement: Python 3.11+.
Supported LLM providers include OpenAI, Anthropic, Azure OpenAI, BytePlus, and more. You bring your own API keys, you control your data.
Guided Learning: Structured Without Being Rigid
The Guided Learning feature turns your uploaded materials into visual learning journeys. DeepTutor:
- Analyzes your materials and designs a multi-step learning plan
- Generates interactive pages for each knowledge point
- Lets you discuss alongside each step with full context
It's like having a curriculum designer who actually knows what you already understand and what you need next.
Who Should Pay Attention
- Developers building EdTech products — The architecture (agent-native, two-layer plugin model) is a blueprint for building adaptive learning systems
- Self-learners and students — Free, open-source, self-hosted tutoring that actually remembers you
- Teams training on internal knowledge — Upload your docs, create TutorBots that onboard new team members
- Researchers in AI education — 10k stars and active development make this a legitimate research platform
What's Coming
The v1.0.x releases are shipping fast — v1.0.0 was April 4, v1.0.2 already landed by April 11. The team is iterating on:
- Search consolidation with SearXNG fallback
- Visualization capabilities (Chart.js/SVG pipeline)
- Cross-platform compatibility improvements
- Full i18n coverage (Chinese already supported, more languages coming)
Why This Matters for the AI Ecosystem
We're in the middle of a shift from "AI as a tool" to "AI as an agent." DeepTutor is one of the first open-source projects that truly embodies this for education. It's not a chatbot that tutors — it's an agent framework purpose-built for learning.
The persistent memory, autonomous TutorBots, and unified workspace all point to a future where AI learning companions actually know you. That's the promise that's been hyped for years. DeepTutor is shipping it.
Links:
- GitHub: HKUDS/DeepTutor
- License: Apache 2.0
- Discord community available for support
FAQ
Is DeepTutor free to use? Yes. DeepTutor is fully open-source under the Apache 2.0 license. You'll need to provide your own LLM API keys (OpenAI, Anthropic, etc.), but the platform itself costs nothing.
Can I run DeepTutor without sending my data to third parties? DeepTutor runs locally on your machine. Your documents and learning data stay on your infrastructure. The only external calls are to your configured LLM and embedding providers.
What's the difference between DeepTutor and using ChatGPT for learning? ChatGPT is stateless — it doesn't remember your learning history, adapt to your level, or build a profile over time. DeepTutor maintains persistent memory across sessions, creates autonomous TutorBots that evolve with you, and organizes your knowledge into structured learning journeys.
Does DeepTutor support languages other than English? Chinese localization is fully supported. The team is actively working on broader i18n coverage. The v1.0.0 release includes full Chinese interface support.
What do I need to run DeepTutor? Python 3.11+, an LLM API key (OpenAI, Anthropic, or others), and for knowledge base features, an embedding API key. Docker images are also available for containerized deployment.