← The AI Hype Audit — all 268 verdicts

REAL Real: OpenMAIC is a genuine Tsinghua open-source project (31.7k stars, MIT) that builds AI classrooms from PDFs; only the 'comment CLASS' DM gate and 'pay later' line are invented Facebook · Sep 2026

The claim24-second reel over the OpenMAIC GitHub README (29k stars): "Turn PDFs into AI classrooms… AI will answer anything. It won't teach you… Drop in any topic/pdf… Includes: quizzes, interactive AI teacher + students… Students don't watch. They pull the levers… Teacher can call on you… Self-host the whole thing, bring your own models… Comment 'CLASS' and I'll DM you this repo… Grab it or pay later." Sep 5, 2026.

The tool in the reel exists and does what the captions say. THU-MAIC/OpenMAIC is a public GitHub repo created Mar 11 2026, showing 31,763 stars, 5,220 forks and an MIT license as of Sep 5 2026 (the reel's '29k' is simply a few days old; last push Sep 4 matches the '13 hours ago' commits). v1.0.0 shipped Aug 27 2026, adding a chat-driven 'Pro workbench' on top of the classic one-click generator. THU-MAIC is Tsinghua University's MAIC lab; the underlying research is 'From MOOC to MAIC' (arXiv 2409.03512, Sep 5 2024; JCST vol. 41 no. 1, Jan 2026, 500+ students, 100k+ learning records) and MAIC-UI (arXiv 2604.25806, Apr 28 2026). The README confirms the pipeline (Outline stage then Scenes stage), inputs (topic, PDF, Word, PPTX, spreadsheets, text, images, audio, video), exports (.pptx, interactive HTML, classroom ZIP, MP4 via an optional Chromium/FFmpeg render container), providers (OpenAI, Anthropic, Azure, Gemini, DeepSeek, Ollama and more) and OpenClaw chat integration (Feishu, Slack, Telegram, Discord, WhatsApp). Self-hosting needs Node 22.19+, pnpm 10+, one LLM key; Postgres is optional. A free hosted demo runs at open.maic.chat. 'Comment CLASS and I'll DM you' is a standard ManyChat-style comment-keyword auto-DM (replient guide, updated Apr 23 2026) for list-building; the repo is one search away. 'Grab it or pay later' has no receipt: the project moved from AGPL to MIT on Jun 29 2026 (v0.3.0), the opposite direction of going paid.

What holds up

  • GitHub API Sep 5 2026: THU-MAIC/OpenMAIC, 31,763 stars, 5,220 forks, 223 open issues, MIT, TypeScript, created Mar 11 2026, pushed Sep 4 2026.
  • Releases page: v0.1.0 Mar 26 2026 through v1.0.0 Aug 27 2026; v0.3.0 (Jun 29 2026) relicensed AGPL-3.0 to MIT; v0.2.0 (Apr 19 2026) added Deep Interactive Mode (3D, simulation, games); v0.3.1 (Jul 21 2026) added MP4 export.
  • README: two-stage Outline then Scenes pipeline; inputs include PDF/PPTX/audio/video; exports .pptx/.html/ZIP (+MP4 via render-service); Ollama supported; OpenClaw skill connects Feishu/Slack/Telegram; live demo open.maic.chat; FunASR local speech docs exist.
  • Tsinghua receipts: arXiv 2409.03512 (Sep 5 2024, 33 authors) and JCST 41(1):394-414 (Jan 2026); THU-MAIC GitHub org created Mar 18 2024; MAIC-UI arXiv 2604.25806 (Apr 28 2026) reports a ~3-month classroom pilot (Nov 2025 to Feb 2026).
  • Cost receipt: topaiproduct.com (Mar 18 2026) reports a full course typically costs under $2 in API fees; allclaw.org (Mar 19 2026) flags high API costs with premium models in long sessions and recommends Gemini Flash.

What doesn’t

  • 'Comment CLASS and I'll DM you this repo' is a ManyChat-style auto-DM funnel that harvests engagement and DM opt-ins for a public repo you can reach by typing 'OpenMAIC' into GitHub.
  • 'Grab it or pay later' is invented urgency: no roadmap, release note, or Tsinghua statement says OpenMAIC will become paid, and the license got MORE open (AGPL to MIT, Jun 29 2026).
  • 'One click' is true only for the hosted demo or once you have already installed Node, pnpm, and an LLM API key; a non-technical owner will not get a classroom from a PDF in one click.
  • Lesson quality is only as good as the model and the PDF parse: the project's own community notes cite poor parsing on complex documents (tables, equations) without the MinerU add-on, and the peer-reviewed evidence is from university courses, not staff manuals.
  • Voice, TTS, and video export add real cost and setup (TTS providers, optional Chromium/FFmpeg container); the reel shows the finished product, not the plumbing.

The catch

The software is real, free, and better-documented than most viral AI tools; the reel's only fiction is the scarcity. The practical catch is that generated lessons need a human read-through before you put them in front of staff or kids, because the AI teacher will confidently narrate whatever the model got wrong.

How to actually do it

  • Skip the comment gate: go straight to github.com/THU-MAIC/OpenMAIC or try the free hosted demo at open.maic.chat with one of your own PDFs.
  • Pick one real use case first (e.g., a 10-page employee handbook or safety SOP) and generate a class with a cheap model like Gemini Flash; budget roughly $1-2 per course in API fees.
  • Read every generated slide and quiz answer against the source PDF before anyone trains on it; export to .pptx so a human can fix errors in PowerPoint or Google Slides.
  • Only self-host if someone on your team can run Node.js/pnpm; otherwise use the hosted demo or have your web/IT person deploy it to Vercel or Docker.

A small business can realistically turn an existing PDF manual into a reviewed, narrated training module for a couple of dollars in API spend, but the reviewing step is not optional.

Confidence
High
Posted by
a creator page

See the original claim →

We test hype for free. We build the real thing for a living.

Thirty minutes, no pitch — and you'll leave with something useful either way.

Book a call with Todd or start with the free Business Checkup →

The Verdict Weekly

Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.

© Schreier Group · schreiergroup.com · See a wild AI claim? Drop it here and we'll test it.