Open-Source AI Education Tools Audit

Five dimensions, 1–5 each

Scores reflect the public state of each repository on 2026-10-19: stars, forks, last push date, README/docs depth, install path, and license. Re-scored from the 2026-10-18 edition: llm-course drops a maintenance point (no push in 8 months); Polymath and Aegis Graph gain a docs point each after their README overhauls. Deployment was judged from documented install flows, not local runs — where we couldn't verify, we scored conservatively and said so. Per our independence charter, ACLAS tools get no advantage for being ours; if anything, we graded them harder.

The table

DeepTutor

HKUDS · University of Hong Kong

Verdict — The benchmark. pip install, Docker, multi-agent tutoring, updated daily by a university lab — last push 2026-10-18, still the most actively developed tool on this list. If you deploy one tool from this list, deploy this one.

Generative AI for Beginners

Microsoft

Verdict — The best free curriculum for learning AI itself — 21 lessons, maintained by Microsoft, pushed 2026-10-17. Scores 4 on teaching fit only because it teaches AI, rather than being a tool for teaching.

Anki

Ankitects

Verdict — Not an AI tool — the reference point. Two decades of spaced repetition, pushed 2026-10-16; every AI flashcard generator is measured against it. Included so the AI-native tools have something honest to beat.

llm-course

Maxime Labonne

Verdict — The standard self-study roadmap into LLMs, with runnable notebooks at every stop. But no push since 2026-02-05 — eight months quiet. The content is stable (fundamentals don’t rot fast), so we drop maintenance to 2: a solo-maintainer project this large needs a succession plan, not just stable content.

Neuro-Edu SDK

ACLAS

Verdict — Our cognitive simulation framework. Strong concept — on-device neural kernels, multi-agent social learning — but docs and onboarding trail the leaders. We’re working on it; the issues page is open.

[AI]lice

LMDDC Luxembourg

Verdict — Tiny but philosophically closest to us: RAG grounded in your course materials, every answer cited, Moodle sync. If you teach and self-host, this deserves more stars than it has.

Aegis Graph

ACLAS

Verdict — Credential verification via agentic GraphRAG — the hardest deployment on this list, and the narrowest teaching fit. Docs up from 3 to 4 after the 2026-10-19 overhaul (rewritten Quick Start, fraud-blacklist v2.0 documentation, Python examples). Powerful where it applies; honest about where it doesn’t.

Open TutorAI

Open TutorAI

Verdict — Ambitious — LLM tutoring with 3D avatars and an arXiv paper. But the repo has been quiet since June 2026. Promising prototype, not yet a platform.

SWOT

ACLAS

Verdict — A tiny utility — identifies academic email domains for discount approval. Does one thing well; scores low on teaching fit because it isn’t a teaching tool and doesn’t pretend to be.

Polymath

ACLAS

Verdict — Our higher-education knowledge graph — courses, prerequisites, ISCED-F/CIP codes. Docs up from 2 to 3 after the 2026-10-19 README overhaul (corrected data examples, Python quickstart, internal memos removed) — but zero stars and zero forks mean nobody has found it yet. The asset is real; the audience is not.

Disagree with a score?

Send evidence to [email protected] — a better doc, a simpler install, a commit we missed. Verified corrections update the table with attribution, per our independence charter.