The learning platform that remembers
Learns with you,for you.
When training adapts to the learner, retention rises and support burden falls. Sudar makes that practical, an open platform where authoring, delivery, and intelligence work as one system.
Start from a document, URL, or brief in Studio. Sudar carries context forward session to session, what someone struggled with, what they've mastered, when they need a nudge, and shapes what comes next.
Open source · Self-host free · Try Studio
A moment on the job
“I paused the video.”Sudar already knew why.
Sudar
You mixed up the escalation steps on Tuesday. Want the short version for this scene?
Marcus
Yes. Walk me through it.
Session 4 · Video modality
Context carried forward
$370B
spent on corporate training annually
Much of it never shows up on the job.
15%
average LMS course completion rate
A number teams have learned to expect.
70%
of new knowledge forgotten within 24h
Ebbinghaus measured this in 1885.
Corporate learning is broken.
Not for lack of effort.
Most LMS platforms (Moodle, Canvas, Blackboard) were built to host content and record completions. The same module goes to every employee, whether they are new to the topic or not, and whether they learn better from text, audio, or practice. Last session's struggles rarely shape what they see next.
Learning science has been clearer for decades: review at intervals, offer more than one format, adjust difficulty to prior knowledge, give feedback while people work. Those ideas are easy to describe and hard to find in the tools most companies buy.
Training budgets grow; transfer often does not. Teams that need real skills get compliance checklists. L&D leads who want better outcomes spend their time fighting software instead of improving courses.
“We built Sudar because the category needed to catch up. The research has been there for a long time.”
Studio. Intelligence. Learn.
One learner model. Three surfaces.
Studio, Intelligence, and Learn share one database. What you publish in Studio and what learners do in Learn both feed the same profile, so recommendations stay consistent.
Sudar Studio
From a document or URL to a published course in under ten minutes.
Upload a PDF, paste a URL, or describe a topic. Studio drafts structure and copy across 14 templates so you can edit, approve, and publish without a separate design team.
- AI course generation from any document
- 14 professional visual templates
- SCORM & LMS export
- Org-wide governance & approval flows
Sudar Intelligence
The AI brain shared by Studio and Learn.
A FastAPI service shared by Studio and Learn. It maintains the Digital Learner Twin (a persistent profile per person), scores modality fit and skill gaps, and returns next-best-action recommendations as events arrive.
- Digital Learner Twin (persistent learner model)
- Adaptive content sequencing engine
- AI tutor with governed longitudinal memory (learner + org cadence)
- Real-time next-best-action inference
Sudar Learn
Every learner gets a tutor, a path, and seven ways to learn.
Where learners take courses. They can switch between text, video, audio, mind map, flashcards, short-form feed, or game-style practice. Tutor Sudar answers questions, nudges when someone is stuck, and uses prior sessions as context.
- 7 learning modalities
- Tutor Sudar with session memory
- Mobile-first layout
- Progress and skill tracking
Built in the open. Documented here.
Walk through animated wireframes for authoring, delivery, integrations, and ops, or jump straight into the live apps.
studio
Document & URL to course
Upload PDF/DOCX or paste a URL; AI drafts outline and modules with RAG over your sources.
Walkthrough →learn
Seven learning modalities
One authored course; learners switch Read, Listen, Watch, Map, Cards, Feed, Play.
Walkthrough →learn
Tutor Sudar & longitudinal memory
RAG over course content, session memory, proactive nudges, and governed LLM cadence.
Walkthrough →integrations
Sudar MCP (Cursor & ChatGPT)
stdio package, Cloudflare remote OAuth worker, creator tools for course generation.
Walkthrough →learn
Notification sounds
Optional chimes for generation complete, tutor reply, toasts, and celebrations.
Walkthrough →learn
30+ UI locales & multilingual TTS
Cookie-driven UI language, content language prefs, RTL fonts, and localized tutor prompts.
Walkthrough →A profile that updates as people actually learn.
The Digital Learner Twin is Sudar's per-learner record. It goes beyond a completion checkbox: modality use, time on task, quiz results, tutor exchanges, and drop-off points accumulate over time.
Learners never see a raw dashboard of scores. The twin sits behind routing (which modality to suggest), when the tutor should check in, which module to surface next, and when someone looks ready for an assessment.
The goal is practical: fewer generic paths, more responses that fit what this person has already done.
Modality Affinity
Does this learner absorb video better than text? Prefer flashcards over long-form? Sudar knows.
Engagement Patterns
Session duration, replay rate, drop-off points, time-to-completion. Every interaction is a signal.
Skill Graph
What has been mastered? Where are the gaps? What prerequisite knowledge is missing?
Cognitive Load Index
Is the learner being overwhelmed or under-challenged? Content complexity adjusts in real time.
Session Memory
Tutor Sudar can refer back to earlier questions, misconceptions, and progress instead of treating each visit as a first meeting.
Next Best Action
At the end of every session, Intelligence computes the single most valuable next step for this learner.
Twin data stays in infrastructure you control, your database, your region, your retention policies. Sudar is built for data sovereignty: self-host in your tenant, choose your AI providers, and keep learner records where your organisation requires them.
Your data. Your infrastructure.
Your learners' rights.
Sudar's reference stack runs on Postgres with tenant isolation, but the principle is broader: learning records should live where you choose, your cloud account, your VPC, your region, with policies your legal team can defend.
Safety
AI that stays inside the course, with guardrails operators can see and tune.
- Sensitive-input checks on tutor and generation routes before content reaches a model.
- Tutor Sudar is RAG-bound to published course material, not the open web.
- Org governance in Studio: memory cadence, personalization limits, and approval flows.
- Proactive nudges are designed to help, not pressure; configurable per deployment.
Privacy
Tenant-scoped by default. Learner context is collected for learning, not resale.
- Organisation-scoped data, no cross-tenant reads in the reference architecture.
- Learners can review surfaced context, adjust preferences, and opt out of tutor memory.
- You choose cloud inference or private/self-hosted model endpoints.
- Sudar does not sell learner profiles; deployers control subprocessors and API keys.
Security
Hardening you can audit, especially when you self-host.
- Role-aware access and row-level isolation in the reference Postgres stack.
- Fail-closed scheduled jobs, dedicated signing secrets, and SSRF-resistant ingestion.
- SCORM, embed, and media proxies enforce ownership before privileged access.
- Technical trust pack in the repo: threat model, data flows, and operator checklists.
Compliance & rights
Built for teams who must answer GDPR, FERPA, and UK GDPR questions.
- Deployer remains data controller; Sudar supplies software and documented flows.
- Support for access, correction, export, and erasure workflows (policy-driven retention).
- Subprocessor transparency, hosting, email, and model vendors are your choices.
- Governance surfaces in Studio link operators to the evidence procurement needs.
Long-term alignment
Learning data exists to serve the learner, not to enrich a vendor's model or ad business.
Sudar does not claim ownership of learner profiles, tutor conversations, or progress records. The Digital Learner Twin is a tool for better teaching, not a data asset for Sudar or your organisation to monetise without the learner's knowledge and consent.
We publish the trust pack and source code so institutions can verify behaviour instead of trusting marketing slides. Product direction stays anchored to learner agency: portable records, minimal necessary inference, and governance defaults that favour privacy over surveillance.
One course. Seven ways to experience it.
Write the course once in Studio. Intelligence can render the same material as text, audio, video, cards, and more without maintaining separate copies for each format.
Grounded in published research.
Mapped to product choices you can inspect.
Durable foundations from Ebbinghaus and Mayer, modern validation from retrieval practice and tutoring meta-analyses, and 2020s AI-era trials — each mapped to a feature you can inspect in the open codebase.
Forgetting Curve
Hermann Ebbinghaus · 1885 · Über das Gedächtnis
Without reinforcement, learners forget most new information within days — a pattern replicated for over a century. Spaced review is the standard countermeasure.
↳ Sudar: Spaced repetition in Flashcards & Adaptive Sequencing
Multimedia Learning
Richard E. Mayer · 2009 · Multimedia Learning, 2nd ed.
People learn more deeply when words and visuals work together than from text alone. Learners differ in which channel combinations help them encode material.
↳ Sudar: Seven adaptive modalities per course
Testing Effect
Roediger & Karpicke · 2006 · Psychological Science, 17(3)
Retrieval practice — answering questions, not just re-reading — strengthens long-term retention more than passive review alone.
↳ Sudar: In-module quizzes, flashcards, and struggle signals in the Digital Learner Twin
Intelligent Tutoring Systems
Kurt VanLehn · 2011 · Educational Psychologist, 46(4)
Meta-analysis shows adaptive tutoring and ITS approaches outperform fixed classroom instruction — the gap mainstream LMS products still leave open.
↳ Sudar: Adaptive sequencing, Next Best Action, and the AI tutor sidebar
AI-Augmented Textbooks
LearnLM Team (Google) · 2025 · arXiv:2509.13348
Personalised, multimodal views of the same source material (text, slides, audio, mind maps) with formative checks improved learning in a randomised trial.
↳ Sudar: Author once, deliver in text, video, audio, mind map, flashcards, and more
Memory-Aware AI Tutoring
Liu et al. (AgentTutor) · 2026 · arXiv:2601.04219
Multi-turn tutoring with knowledge memory and strategy adjustment beats stateless chat — yet most LLM tutors still reset every session.
↳ Sudar: Cross-session tutor memory in ai_tutor_context and consent-governed learner model updates
The best learning tools
should not cost a fortune.
Sudar is Apache-2.0 and designed to run on infrastructure you choose: common free tiers for the Next.js apps and Intelligence, plus a Postgres-compatible database in your account or VPC. Many teams can host for $0 before they outgrow hobby limits.
You can read the code that scores learners, routes modalities, and powers the tutor. Fork it, audit it, or patch it for your org without negotiating a proprietary roadmap.
Manifesto
Learners differ. Defaulting everyone to the same path is a design choice, not a law of nature.
The research on how people learn is extensive. Most products still ship as if it were not.
Source code you can read beats black-box claims about personalization.
A tutor that keeps context across sessions behaves differently from a one-off chat window.
Corporate training works better when it adapts to the person, not only the compliance deadline.
Your learners deserve better
than what they’re getting.
Open Sudar Studio (free, no card). Publish a first course from a document or URL and walk through it in Learn to see how the tutor and modalities behave.