DeepTutor is an open-source, self-hostable Agent-Native learning assistant combining persistent memory, TutorBot agents, five learning modes, and a RAG knowledge base. This series takes you from zero to fully operational.
DeepTutor is an open-source Agent-Native personalized learning assistant developed by HKUDS at the University of Hong Kong. This article breaks down its fundamental differences from ChatGPT and explains why persistent memory is the core of any serious AI learning tool.
A complete DeepTutor installation guide covering both Docker deployment and native local installation, including LLM backend configuration, common error handling, and initial setup steps before your first session.
DeepTutor's power lies in five distinct learning modes within a single interface. This article breaks down Chat, Deep Solve, Quiz, Research, and Math Animator—when to use each, how to operate them, and how to combine them for maximum effect.
DeepTutor's TutorBot is an AI tutor agent with personality and isolated memory. This article covers how to design a TutorBot's teaching style, configure proactive reminders, and connect it to Telegram, Discord, and other platforms so it fits naturally into your learning rhythm.
DeepTutor's Knowledge Base combines LightRAG semantic indexing to help AI genuinely understand your PDFs, notes, and technical documents. This article covers how to build a knowledge base, manage your data, optimize retrieval quality, and get the most out of RAG in learning and research workflows.
The first five articles built the map of DeepTutor. This one is the navigation. Three users from different backgrounds—students, engineers, and researchers—each showing how to integrate DeepTutor's five modes, TutorBot, and Knowledge Base into real learning and work workflows.
All three claim to make AI-assisted development more productive—but they're solving fundamentally different problems. This post maps out their core purposes, ideal use cases, and how to combine them effectively.
Gstack is the Claude Code workflow framework built by Y Combinator CEO Garry Tan — not more prompts, but a way to put AI in the right cognitive mode. This series walks you through everything from installation to real-world development scenarios in six posts.
When you let Claude handle product decisions, architecture design, and code review all in the same conversation — everything comes out mediocre. Gstack's core concept: give AI the right cognitive mode, not a universal prompt.