mrkeyoor.com_
Tue 29 Sept 18:28 UTC
AI Toolsevaluationupdated 26 Aug 2026

OpenMAIC review

OpenMAIC is an English-documented, open-source classroom generator that turns a topic or uploaded material into slides, quizzes, simulations, project work, and conversations with AI teachers and classmates. It can play lessons in the browser, edit generated scenes, and export presentations, interactive HTML, or video.

+619stars / 7d
Verdict

Our OpenMAIC build passed in 116 seconds and 4,892 of 4,954 tests passed, but installation consumed 2,672 MB and one guarded-import test failed. It is a strong trial candidate for educators who want to edit an AI-built lesson and can review every claim before delivery. Institutions should add real identity, authorization, provider governance, and content approval before treating it as a shared learning system.

We ran it

Lab card: what happened when we ran OpenMAICScreenshot of OpenMAIC (github.com/THU-MAIC/OpenMAIC)
Install✓ · 176s2546 packages · 2958 MB
Build✓ · 162s
Tests✗ · 371s8848 passed · 1 failed · 98 skipped of 8956 (vitest)
Repo3100 files~616,786 lines of source · 116.9 MB · 6 CI workflows · Dockerfile · tests dir

Answers from our run

Does OpenMAIC build from source?

Dependencies installed in 176 seconds (2546 packages), and the build succeeded in 162 seconds. We cloned commit f287542 into a clean Debian container with 3 CPUs and no project-specific setup.

Do OpenMAIC's tests pass?

Not all of them: 8848 of 8956 passed and 1 failed when we ran the project's own test command (vitest). Some failures need services or credentials a bare container does not have.

Who should not use OpenMAIC?

Schools expecting generated lessons to be fact-checked automatically: the README describes generation and editing, not a guarantee of curricular accuracy.

What are the alternatives to OpenMAIC?

Open edX, Moodle, LobeChat. Our OpenMAIC build passed in 116 seconds and 4,892 of 4,954 tests passed, but installation consumed 2,672 MB and one guarded-import test failed.

Setup3/5Build passed, but 2,569 packages used 2,672 MB
Docs5/5Providers, Docker, persistence, exports, and security are explicit
Community5/520,932 stars, 202 issues and PRs, pushed August 26
Maturity3/5Fast-moving v0.3.2 project with one failing test

Discussed on

  1. hnOpen Multi-Agent Interactive Classroom3 points

Who it’s for

Educators who will inspect and revise AI-generated lessons before students see them.
Training teams that want an editable classroom experience rather than a plain chatbot answer.
Developers who need several model, speech, search, image, and local-provider choices.
Self-hosters prepared to operate a Next.js monorepo, optional PostgreSQL storage, and media services.

Who it’s NOT for

Schools expecting generated lessons to be fact-checked automatically: the README describes generation and editing, not a guarantee of curricular accuracy.
Public multi-user deployments relying only on the documented persistence token: the README says it is visible in browser JavaScript and provides no user isolation.
Small servers or thin developer laptops: our install added 2,569 packages and occupied 2,672 MB before generated assets.
Teams seeking a finished identity and authorization system: the shared ACCESS_CODE is site-level protection, while production persistence requires real session verification.
Buyers who need a settled product surface: the repository was created in March 2026 and reached v0.3.2 in August.

Setup reality

Our commit 0088034 checkout installed 2,569 packages in 133 seconds and used 2,672 MB. The build passed in 116 seconds. Tests failed after 87 seconds: 4,892 passed, 1 failed, and 61 were skipped out of 4,954. The excerpt locates the failure in a guarded-import test but does not show the failing assertion.

At least one LLM provider is needed, unless a compatible local service such as Ollama or Lemonade supplies it. Speech, images, video, search, document parsing, and Bedrock add separate credentials or services; PostgreSQL is optional for server persistence.

The 108.5 MB monorepo had 2,310 files, about 429,513 source lines, 5 CI workflows, a Dockerfile, compose configuration, and tests. Node 20 and pnpm 10 are documented prerequisites.

OpenMAIC generates a classroom, not a single answer

OpenMAIC takes a subject or uploaded material and creates a staged lesson. AI teachers and classmates can speak, write on a whiteboard, discuss the topic, present slides, ask quiz questions, and run interactive or project-based activities. The browser handles playback and learner interaction, while an editor lets the author adjust outlines and scenes instead of accepting the first generation unchanged.

That format solves a different problem from a tutoring chatbot. A chat responds turn by turn; OpenMAIC assembles an experience with an order, roles, media, and assessable moments. Exports include editable PowerPoint files, self-contained interactive HTML, classroom archives, and MP4 video. An educator can reuse the output outside the live application, subject to checking what each export preserves.

What happened when we ran it

Our sandbox cloned commit 0088034 with 3 CPUs and 8 GB of RAM. The 108.5 MB checkout contained 2,310 files and about 429,513 lines of source. pnpm install succeeded in 133 seconds, adding 2,569 packages and occupying 2,672 MB. That is a large local footprint before model caches, PostgreSQL, generated images, audio, or video assets.

The production build completed successfully in 116 seconds. Vitest then ran for 87 seconds and returned exit code 1: 4,892 tests passed, 1 failed, and 61 were skipped out of 4,954. The supplied excerpt points to a test that checks blocked imports across guarded paths. It does not include the assertion message, so we cannot say whether the fault was a parser rule, fixture, path, or environment.

The monorepo has 5 CI workflow files, a Dockerfile, compose configuration, and a tests directory. Node 20 and pnpm 10 are the stated minimums. A one-command container path exists, but the 2,672 MB dependency result makes resource budgeting worthwhile even when Docker hides the package installation steps.

One model key starts it, while rich lessons need more services

The basic configuration needs at least one language-model provider. OpenMAIC supports hosted vendors, Amazon Bedrock, OpenAI-compatible endpoints, and local options such as Ollama and Lemonade. Per-stage routing can send different parts of generation to different models. That flexibility is useful for cost and data policy, but it creates a matrix of model names, credentials, rate limits, and output differences.

A complete media lesson may also use image generation, text to speech, speech recognition, search, video generation, or document parsing. Some providers reuse a language-model account; others need another endpoint or key. Local FunASR and Lemonade routes avoid hosted credentials but bring their own Python, model, hardware, and service requirements. One click describes the learner interaction better than the operator setup.

Provider selection affects the lesson itself. Different models can produce different outlines, quiz distractors, citations, and scene instructions from the same document. Pin the model and prompt configuration for assessed material, save the source used, and run a review step after generation. A polished voice and animated whiteboard can make a false claim feel more authoritative than plain text.

PostgreSQL persistence is documented as trusted-network only

By default, course data can live in the browser. The optional server-persistence compose profile adds PostgreSQL and an embedded HTTP endpoint so sessions and course documents can move to server storage. Device-scoped values still remain in the browser, and existing browser courses migrate lazily as they are accessed. This is enough for a controlled single-user or trusted-network deployment.

The README is unusually direct about its development token. NEXT_PUBLIC_PERSISTENCE_TOKEN is compiled into public JavaScript, so anyone who can load the application can extract it. The token offers no confidentiality or learner isolation, and a caller can choose a learner key. Production use requires server-verified sessions that derive the learner partition from identity, plus authorization rules for documents, merges, and administration.

A separate ACCESS_CODE can place a shared password in front of the site and API routes. That can deter casual access, but it is not a school identity system with students, teachers, classes, and revocation. Open edX or Moodle starts much closer to those institutional requirements, though neither aims to generate this kind of multi-agent lesson on demand.

Version 0.3.2 adds storage and export machinery quickly

GitHub recorded 20,932 stars, 202 combined open issues and pull requests, and a last push on August 26, 2026. The repository was created on March 11, 2026, and v0.3.2 was published on August 14. That is rapid adoption and code movement over roughly five months. It is also a young operational surface despite the high star count.

The v0.3.2 notes cover deterministic quiz and project covers for video, interactive HTML capture, CPU profiles, parallel rendering, PostgreSQL document storage, incremental saves, an asset registry, new locales, and additional model providers. Fixes span binary persistence responses, chat migration, access-code provider loading, Windows image startup, international fonts, and release packaging. The breadth explains both the product's appeal and the need for controlled upgrades.

The root is MIT-licensed, after a June 2026 change from AGPL, while bundled workspace packages retain their own licenses. Redistributors need to preserve those component terms. The release notes also show that storage contracts and SDK packages are still being separated from the application, so extension points are improving while their ownership is still moving.

OpenMAIC earns a trial when a teacher or training designer wants something more structured than chat and will edit the result. The passing build and 4,892 passing tests are encouraging. The 2,569-package install, single failure, public persistence token, and number of optional providers argue for a sandbox first. Use it as an authoring instrument today; build the missing institutional controls before students depend on it.

Alternatives

ProjectWhat it isPick it when
Open edXA large learning-management platform for authored courses, cohorts, assessment, and delivery.pick this instead when governed course operations and human-authored material matter more than instant AI lesson generation.
MoodleA long-running open-source learning platform with courses, roles, grading, and plugins.pick this instead when institutional identity, enrollment, grading, and administration are the core requirements.
LobeChat gh↗A self-hosted AI chat interface with many providers, knowledge bases, and agents.pick this instead when conversational access to models matters more than generated classroom scenes and playback.

What people are saying

  1. [github-trending] THU-MAIC/OpenMAIC

Sources

  1. OpenMAIC README
  2. OpenMAIC repository facts
  3. OpenMAIC v0.3.2 release
  4. OpenMAIC license

More ai tools reviews

voltagent · InferenceX · Bonsai-demo · qwen-audio-agent · wechat-intelligence-hub · dlss5-visual-enhancer · the whole board →