An AI tutor that remembers you between courses

What it does

StudyMate is a learning workflow and plugin (primarily for DeepSeek Harness, with Antigravity and Codex/ChatGPT variants) aimed at math and CS self-study. It builds a course for you: a syllabus roadmap with knowledge points ordered by prerequisite dependencies, textbook-style lesson pages with figures, exercises, and hands-on projects. Progress lives in files under a workspace (~/StudyMate by default), so every new conversation picks up where the last one left off.

The interesting bit

The memory is cross-subject: it records your level, which explanations worked, and where you got stuck — so starting a new course doesn’t mean re-introducing yourself to the AI. It also organizes five roles (research, image sourcing, course design, teaching, assessment) that get delegated to sub-agents when the host supports it.

Key highlights

• Knowledge-point roadmap with dependency ordering, node status coloring, and expandable lesson cards
• Course types per node: concept, hands-on, or lab
• “Test me” mode generates exercises and verifies answers against runnable evidence
• Adapts teaching style on feedback (“too easy” / “didn’t get it”) and logs the preference
• Can ingest your own textbook, converting it to Markdown as the reference corpus
• Fully open source, MIT licensed

Caveats

• The author openly admits the codebase was largely “vibed out” — bloated docs, defensive code, messy modules — and is being manually reworked file by file
• Codex/ChatGPT and Antigravity support is frozen at v1.0.0; only the DSH side gets updates
• The independent web runtime was moved off main to a branch pending engine adaptation

Verdict

Worth a look if you self-study math or CS and are tired of chat sessions that forget everything by Tuesday. If you’re not in the DSH ecosystem, the frozen v1.0.0 plugin support limits the appeal.