AI-native LMS · open source
Courses built from your docs. Every claim cited.
AI drafts lessons from your material. You review every change before publishing.
The CEO uploads policy and style guide
Start with the files your team already uses.
- Upload PDF, Markdown, DOCX or plain text.
- Add images and a style guide for the course design.
- Keep the policy as the source for lessons.
CEO: Upload policy PDF, Brand style guide. Extract PDF text → Cited source context → LLM reads the policy. § 4 · Customer data: Delete customer exports after 30 days.
AI asks about the organisation
The CEO defines the audience, duration, seniority and experience level. These answers become the course brief.
- Choose departments and roles.
- Set course length, seniority and experience level.
- Review the brief before the course is drafted.
How many departments need this training? 3: Management, HR and Sales. How long should the course take? 30 minutes, split into short lessons. What seniority levels should it cover? Team leaders and individual contributors. What is their experience level? New starters and experienced employees.
One policy, distinct branded academies
Management, HR and Sales each get their own course design and domain, shaped by the same policy and the organisation’s style guide.
- Adapt the content to each department.
- Give each academy its own design and domain.
- Keep tenant data separate.
Management: Decisions and accountability. management.example.test. HR: Employee data and access. hr.example.test. Sales: Customer data and consent. sales.example.test
Learners finish and certify
Interactive lessons lead to an exam. Passing it generates a PDF certificate.
- Work through lessons and interactive exercises.
- Take the final exam.
- Download the PDF certificate.
Your organisation’s academy. Interactive lesson → Knowledge check → Final exam → Certificate of completion PDF v1.
The policy owner tracks completion
See who has completed the required training and who still needs to take it.
- See who finished and who still needs training.
- Check progress by department and course version.
- Review exam results and certificates.
Management learner: Completed. HR learner: Completed. Sales learner: Not completed
One year later: policy v2 changes
In a source-linked course, ulams finds affected lessons and questions and prepares cited edits for the owner to review and approve.
- Upload the revised policy.
- Find the lessons and questions affected by the change.
- Review and approve the update before publishing.
One year later. § 4 · Customer data: Delete customer exports after 30 days. → Delete customer exports after 14 days.. Handling customer data, Final exam. Owner reviews before publishing.
People renew their knowledge
Prepare updates for every affected team course. After approval, notify learners, expire old certificates, assign an update or new course and exam, then issue new PDFs.
- Notify learners and retire old certificates.
- Small change: take a new exam.
- Major change: complete the updated course.
§ 4 changed: data retention is now 14 days. No longer valid. Small change → update & exam. Major change → course & exam. New certificate PDF v2.
live demos
Six academies, one platform
Each demo is a separate tenant with its own theme, course and data, running on the same ulams install. Open it as a learner or step into its admin.
Gravity Lab
A real-data solar system you learn inside.
- Style
- 3D simulation
- Lessons
- 12
- Topics
- 54
- Price
- Free
Poland, Measured
35 years of Poland in cited public data, in English and Polish.
- Style
- Map and charts
- Lessons
- 12
- Topics
- 48
- Price
- Free
The Scottish Book
Ulam, the Lwów School and the Scottish Book, with five live interactives.
- Style
- Mathematics and history
- Lessons
- 11
- Topics
- 47
- Price
- Free
Coffee Academy
A slow, magazine-style course from seed to cup: film, podcast, interactive diagrams, a quiz and a final project.
- Style
- Editorial · self-paced
- Lessons
- 6
- Topics
- 19
On-Call
Incident command for platform engineers: a 4-week cohort with an outage simulator, a live drill and a reviewed postmortem.
- Style
- Cohort · dark pro tool
- Lessons
- 5
- Topics
- 16
Night Sky Explorers
Seven short missions to the stars with Orbi the robot guide, badges and a printable diploma.
- Style
- Gamified · kids 10–14
- Lessons
- 7
- Topics
- 16
- Price
- Free
Demos reset every hour. No sign-up: you are logged in automatically.
The thinking behind the name
Knowledge grows when we build it together.
In the cafés of Lwów, Stanisław Ulam and fellow mathematicians discussed problems and recorded them in a shared notebook: the Scottish Book. Questions, solutions and new ideas grew through collaboration.
Ulam later translated the collection so it could reach more people, inviting readers to contribute corrections and further developments.
That spirit inspires ULAMS: turn your organisation’s documents into shared learning, work with people and agents, and keep improving the knowledge together. Open source, free to use and open to contribution.
Build with us on GitHub(opens in a new tab)run it from anywhere
Manage your academy from the admin panel, Claude, Claude Code or your terminal
ULAMS is a standard LMS. The admin panel is there for courses, people, enrolments and reports. You can also use Claude Code, Claude, Codex, OpenCode, Gemini CLI or another MCP-compatible agent, the ULAMS CLI or the REST API. The studio is a visual workspace for your team.
Six ways to run ulams: the admin panel, Claude Code, Claude with the ulams MCP server, the ulams command line, the REST API and SDK, and the studio.
- Section: Courses
- Section: People
- Section: Enrolments
- Section: Reports
- Course: On-Call. Published | 86
- Course: Gravity Lab. Published | 42
- Course: The Coffee Atlas. Draft | 0
The admin panel is the usual way in: courses, people, enrolments and reports.
- You ask Claude Code: Build a 3-lesson course on incident response from ./runbook.md and publish it
- Claude Code: I'll start a builder run on runbook.md and bring you the outline before anything goes live.
- Tool call: Bash ulams builder start --from runbook.md --defaults --json
- Running: Reading runbook.md, drafting the outline
- Output: {"status":"outline_review","session":"01J9Z3K8M2"}
- Claude Code: Outline ready: 1. Detect and declare, 2. Triage and mitigate, 3. Write the postmortem. Every lesson cites runbook.md. Approve and publish?
- You ask Claude Code: Yes, approve and publish.
- Tool call: Bash ulams builder outline approve 01J9Z3K8M2 --json
- Output: {"status":"apply_review","session":"01J9Z3K8M2"}
- Tool call: Bash ulams builder apply 01J9Z3K8M2 --json
- Output: {"status":"applied","courseId":17}
- Tool call: Bash ulams courses publish 17 --json
- Output: {"id":17,"status":"published"}
- Claude Code: Done. Incident response is live as course 17: 3 lessons, every claim cited to runbook.md.
Describe the course in plain language. Claude Code runs the ulams CLI, shows you the outline, and publishes only after you approve.
- You: Which courses are still drafts? Publish the coffee one.
- Tool call: courses_list status: draft returned 2 courses
- Claude: Two drafts: Learn coffee the slow way and Incident response. I'll publish Learn coffee the slow way. Go ahead?
- You: Yes, publish it.
- Tool call: courses_publish id: 3 returned published
- Claude: Done. Learn coffee the slow way is published. Incident response is still a draft.
Ask Claude, or any MCP-compatible agent, about your academy. It uses the ulams MCP server and asks before it sends anything.
- Command: ulams login --url http://coffee.localhost --demo admin
- Output: Signed in as [email protected]
- Command: ulams courses list --json | jq -r '.data[] | "\(.id) \(.title)"'
- Output: 3 Learn coffee the slow way
- Command: ulams get Course "Learn coffee the slow way" --output yaml > course.yaml
- Comment: # edit the grind-size lesson in course.yaml
- Command: ulams apply --file course.yaml --dry-run
- Output: Course Learn coffee the slow way updated (dry run, nothing written)
- Command: ulams apply --file course.yaml
- Running: Applying 1 change
- Output: Course Learn coffee the slow way updated
Course as code: export a course to YAML, edit it in your editor, review the plan, apply it back.
- Command: curl -s -X POST https://oncall.example.com/api/admin/topics \
- Command, continued: -H "Authorization: Bearer $ULAMS_TOKEN" -H "Content-Type: application/json" \
- Command, continued: -d '{"title":"Reading a page","lesson_id":12,"topicable_type":"EscolaLms\\TopicTypes\\Models\\TopicContent\\RichText","value":"What the alert means."}'
- Output: {"success":true,"data":{"id":88,"title":"Reading a page","lesson_id":12}}
- Command: node -e "import('@ulams/sdk').then(async ({createClient}) => { const ulams = createClient({baseUrl: 'https://oncall.example.com'}); console.log((await ulams.courses.list()).data.length) })"
- Output: 1
Everything the studio does goes through the HTTP API. Plain curl works; so does @ulams/sdk for reading.
- You: Add a quiz after lesson 2 from the runbook.
- Claude: I drafted 5 questions. Each one cites a section of runbook.md, and nothing changes until you approve.
- Proposal: Proposed change: new quiz after Triage and mitigate, 5 questions, all cited Approve | Reject
The studio is the visual course builder: every AI change is a diff you approve.
One platform: the admin your team expects, and every interface they already work in.
Claude and Claude Code are products of Anthropic. ulams is independent and not affiliated with or endorsed by Anthropic.
standards
Delivered anywhere
Standards over lock-in: bring existing packages, track learning anywhere, integrate through the API.
- H5Pinteractive content
- SCORM 1.2 / 2004
- cmi5
- xAPIbuilt-in LRS
- LTI 1.3Platform + Tool
- OpenAPIREST API
- WebhooksComing
self-host
Your servers. Your data.
Run the complete stack with one command on your laptop, Docker server or Kubernetes cluster. Use your own VPS, AWS, Google Cloud or Azure. Your deployment, your data.
- Try it locally(opens in a new tab)
Start an academy with a deterministic AI provider. No paid model calls.
- Docker production(opens in a new tab)
Launch with HTTPS, private credentials and persistent storage.
- Kubernetes YAML(opens in a new tab)
Generate and apply native manifests; the CLI provisions the academy and waits for readiness.
Helm chart coming soon.
$ npm install -g https://github.com/ulams-dev/ulams/releases/download/cli-v0.2.0/ulams-0.2.0.tgz$ ulams stack up✓ academy, Studio, API, admin, H5P and storage$ ulams stack up --mode production \ --domain learn.example.org$ ulams stack up --target kubernetes
CONTACT
Talk with the ULAMS team
Questions about self-hosting, integrations or contributing? Send us a note.
FAQ
Questions before you start
What is ULAMS?
ULAMS is an open-source LMS for organisations, universities and technical teams. Its headless API supports branded learning sites, course authoring and integrations.
Is ULAMS free to self-host?
Yes. ULAMS is MIT licensed. You provide hosting; paid AI provider calls, if enabled, are billed by your chosen provider. H5P runs as a separate service with its own GPL licence.
Can I try the whole stack locally?
Yes. Install the CLI and run ulams stack up with Docker Compose available. The development trial uses deterministic AI responses without paid model calls. Node.js 22.12 or newer is required.
Where can I deploy it?
Use Docker Compose on your own server or native Kubernetes manifests on an existing cluster. A VPS or virtual machine on AWS, Google Cloud or Azure can host the stack. Production needs DNS, HTTPS and persistent storage.
Can agents operate ULAMS?
The CLI and MCP interface let compatible agents work with the platform. You can use Claude Code, Codex, OpenCode or Gemini CLI, or integrate directly through the REST API.
Does AI publish a course automatically?
No. AI prepares a draft with source citations. The author reviews the outline and content, then approves publication. Production starts with AI disabled until you configure a provider.
Can each academy use its own brand and domain?
Yes. Each tenant has its own learning site, theme and isolated database. The initial academy is created by the stack launcher; additional tenants use the platform tenant workflow.
How do I stop it without losing courses?
Run ulams stack down. Data and credentials remain available when you run stack up again in the same installation directory. Only stack down --purge --yes deliberately removes persistent data.