── A personalized learning system, in minutes

Learn anything — with AI that learns you as you go.

Nebular builds a living model of what you understand, session by session. Then any AI agent picks up exactly where you are. Start in one of two ways:

Pick a program, connect your agent, and go.
or
Turn a textbook, notes, or a goal into a path.
Works with any AI agent — Claude · GPT · Gemini · Cursor · OpenCode
── How it works

From your materials, to any agent.

Feed in what you've got. Nebular distills it into one living learner model — which any agent can read from and write back to.

1
What you feed in
materials
Textbook
Intro to ML · 412 pp
Lecture notes
CS229 · 14 weeks
Syllabus
Stanford · Spring
Your goal
“ship an agent”
2
The craft
learner knowledge graph
Learner knowledge graph
alex@stanford · 247 nodes
MasteredIn progressGapUnexplored
ObjectivesPrerequisitesAssessmentsPedagogy
3
Any agent
learn anywhere
◆Claude
◇GPT-4
✦Gemini
⌘Cursor
A
Alex · one learner
Same memory across every agent
── Three ways in

Whether you learn, teach, or hack your own path.

Nebular meets you where you are. Pick a starting point — the learner model is the same underneath.

◐For learners

Learn anything, with any agent.

Pick a program, connect your agent of choice, and study at your level. Your progress is remembered across sessions and tools.

  • ▸Browse 240+ community programs
  • ▸Connect Claude, GPT, Gemini, or Cursor
  • ▸Adaptive practice tuned to your gaps
Browse programs →Or upload your own materials
◆For creators & educators

Turn your knowledge into a learning program.

Drop in a textbook, lecture notes, or a syllabus. Creator Studio extracts objectives, prerequisites, and assessments — ready to share with the world.

  • ▸Auto-generate skill graphs from sources
  • ▸Edit objectives, prereqs, pedagogy
  • ▸Publish or remix community programs
Open Creator Studio →See an example program
⌘For power users & explorers

Build your own learning journey.

Mix three textbooks, a paper, and a podcast into a single path toward whatever you want to master. No curriculum required — just a goal and the agents you trust.

  • ▸Define your own goal and trajectory
  • ▸Remix sources across domains
  • ▸Track mastery in one persistent graph
Start a custom path →See power-user examples
── What you know

See your knowledge take shape.

Every concept you learn becomes part of your personal knowledge map. Strong areas glow. Gaps fade. Misconceptions are flagged, so you can see not just what you missed — but why.

247
Concepts mapped
89
Mastered
12
Gaps to fix
Open your knowledge map →
Alex's knowledge map
Machine Learning · Updated after your last session
live
Linear regressionNeural networksGradient descentLoss functionsAttentionVectorsCalculusTransformersCUDATritonBackprop
Misconception detected
You understand what gradient descent does, but not yet why the step size matters.
Next best step
Review Gradient descent before Backprop.
● Mastered● Learning now● Needs review● Misconception● Not started
── What you can do with it

One learner model. Every agent stays in sync.

Connect any AI agent — Claude, ChatGPT, Gemini, Claude Code, Cursor — to your Nebular learner model over MCP. Each agent reads the same mastery state, picks up where the last one left off, and records evidence back as you work.

Conversational agentClaude / ChatGPT / Gemini
→ Connected via MCP
N
Quick check before we go further. You're standing on a steep slope of the loss surface. You double the step size. What happens to the loss after one step?
It should drop faster — bigger step, faster descent.
N
That's the misconception I want to fix. The step doesn't scale with how fast loss drops — it scales with curvature. Let me build you a one-screen interactive so you can see it overshoot.
Building skills/gradient-descent/intuition/step-size.html
>_Claude Code (local workspace)MCP + .nebular/
→ Pedagogy rules shipped as files in your repo
$ curl -fsSL nebular.live/setup/dynamic-programming | bash
✓ wrote .nebular/AGENT.md (pedagogy contract)
✓ wrote .nebular/PROGRAMME.md (curriculum + skills)
✓ wrote .nebular/LEARNER.md (empty — fill on interview)
✓ registered MCP server nebular-mind
 
$ claude
Read .nebular/AGENT.md, .nebular/PROGRAMME.md, .nebular/LEARNER.md. Running interview first, then session_bootstrap.
Lessons are interactive HTML artifacts under skills/<slug>/, not chat.
 
$
Nebular learner model→Agent context→Lesson + evidence←Evidence back
MCP calls · session #1184 context evidence
✓session_bootstrap:recommended: state designconversational
✓lesson_start:essential: what is the state?conversational
✓skill_neighborhood:prereq: recursion (mastered)claude code
✓evidence_record:state_design · partialconversational
✓evidence_record:memoisation · weakclaude code
✓session_end:summary writtenclaude code
The interface can change. The learner model stays persistent.
Nebular
Built by the Nebular team
"Programme that teaches main concepts on agentic development and acts as a coach in your claude code sessions."
Newest programme

Development with Claude Code

Programme that teaches main concepts on agentic development and acts as a coach in your claude code sessions.

NPublished 5/26/2026
agentic developmentsoftware engineeringvibe coding
Explore programmesView all →

Higgsfield ML position prep kit

Special programme that helps you achieve outcomes required to work as an ML professional in Higgsfield.

mlai engineering
N
Nebular
Built by the Nebular team
▶ Start

69 shades or Rust

Published learning programme on Nebular.

rustweb assembly
N
Nebular
Built by the Nebular team
▶ Start

Deep Knowledge Tracing

Programme on deep knowledge tracing based on the paper by Stanford University, Khan Academy and Google

N
Nebular
Built by the Nebular team
▶ Start

Otter RL introduction for Robotics

Otters are fun creatures. They do strange stuff. Like throwing smokes... or using robotics arms for...

rlml
N
Nebular
Built by the Nebular team
▶ Start

Competitive programming program

Programme is based on Competitive Programming Handbook by Antti Laaksonen

N
Nebular
Built by the Nebular team
▶ Start