DeepTutor is an open-source agent-native personalized learning platform from the HKUDS lab (Hong Kong University of Science). It reached 10,000 GitHub stars in just 39 days following its v1.0.0-beta.1 release. Unlike stateless AI chatbots, DeepTutor creates persistent TutorBots — long-running agents with independent memory systems, customizable teaching philosophies, and dedicated workspaces that evolve over time. Five learning modes (Chat, Deep Solve, Quiz Generation, Deep Research, Math Animator) operate within a single conversation thread, preserving full context when users switch between tools. Users build RAG-ready knowledge bases by uploading PDFs, Markdown, and text files, which TutorBots use to ground responses in personal materials. An evolving learner profile captures study history, knowledge gaps, and individual preferences — shared across all features and bots. The AI Co-Writer Markdown editor treats AI as a first-class collaborator, drawing from knowledge bases and live web sources. Multi-channel deployment supports Telegram, Discord, Slack, WeChat, Feishu, and DingTalk. The agent-native CLI outputs JSON, enabling external AI orchestrators to autonomously configure and operate the platform.