Turn almost anything into a course you actually finish

A repo, an article, a video, a book — one pipeline, one gated course.

Download for desktop Star on GitHub

Local-first
Bring your own key
No account

Windows · macOS · Linux · Android

GitHub repo Web article Bilibili video EPUB book

Parsing source

0%course generated

Answer to light the next balloon

Exams guard each gate — master a lesson and the next one unlocks.

Complex courses become Duolingo-style micro-lessons.

AI tracks mastery at every node — checking in feels easy.

These balloons double as pinballs. Throw one.

Every chapter ends with a mini boss fight.

Drag the balloons. Throw the bot. Scroll — the sky turns.

A real tutor, from first question to mastery

It answers, quizzes you, proposes mastery and schedules the exam.

Start learning: Neural Network Basics

Quick check — what does an activation function do?

It compresses inputs to save memory It adds non-linearity, so nets fit complex boundaries It is a regularizer that prevents overfitting

Correct — mastery updated

Proposal · Mark “Neural Network Basics” mastered + schedule a quiz SRS will schedule it right before you’d forget

Applied — scheduled in SRS ✓

Neural nets 41%

Backprop 91%

Read it aloud, note it, see the map

Highlights become notes; the blackboard grows a concept map.

A neuron first takes a weighted sum: z = w₁x₁ + w₂x₂ + b. Then an activation adds non-linearity: σ(z) = 1 / (1 + e^−z). So even a single layer can learn a curved decision boundary.

Note · re-read the activation formula before the quiz

Blackboard · concept map

σ(z) = 1 / (1 + e^−z)

A study buddy that is yours

Five preset forms, your own artwork, or classic Shimeji.

Five preset forms

Ember / Frost / Moss / Astro / Ink, one tap away.

Bring your own puppet

Import a PNG; it is cut into a rig with all the same physics.

Classic Shimeji

Load Shimeji packs — they walk, climb and get thrown.

The real thing

Three panes, one screen: the map on the left, the tutor in the middle, your notebook on the right.

LookatStudy with the skill map on the left, tutor chat in the middle, notebook on the right

The tutor explaining the Propose → Apply protocol in chat — Propose → Apply: the AI drafts, you decide

A timed chapter exam question with one option selected — Timed questions: walk away and it counts

Why a course, not another tab

“I star a lot of tutorials and finish almost none of them.”

A pile of docs is missing the three things every course has. A path: the map decides what you study today, and locks the rest until you’ve earned it. Feedback: lessons break into knowledge points and your mastery is the weakest one, so nothing slips through. A reason to return: SM-2 brings material back right before you’d forget, and XP, streaks and crowns give tomorrow-you a reason to open it again.

Start finishing things

A six-chapter guide course ships built in — you can click through the whole loop without an API key.

Grab an installer

Windows NSIS · macOS Apple-silicon dmg (unsigned: right-click → Open) · Linux AppImage + deb.

Windows macOS Linux

Android path: a launcher APK installs Termux and runs the same app in Chrome — your data stays on the phone.

Or run from source

Any platform, Node 22 or newer:

git clone https://github.com/Kaiji-Z/LookatStudy.git
cd LookatStudy
npm install
npm run dev:electron

To bring the AI in: Settings → pick one of nineteen providers → paste your key → Test Connection.