Agent-first notes, over MCP

Let the agent write.
You keep the desk.

Connect Claude, ChatGPT, Cursor, or another MCP client. They write quizzes, cards and diagrams into the same notebook you open to study. Native blocks, cue-linked, yours to revoke.

Your agent, your workspacePrivate by defaultRevoke in one click

Concurrency in Go

Text

Agents write quizzes, cards and flows straight into the notebook you already opened. Real blocks, not a scrape.

Heading0:00

Concurrency in Go: goroutines, channels, selects

Quote58:30

“Do not communicate by sharing memory; instead, share memory by communicating.”

Callout12:04

Errors are values. Return them. Do not throw them.

Formula

select { case msg := <-ch: return msg; default: return err }

Code runner

for msg := range ch { handle(msg) }

Image

worker-pool.png, a source image block

Definition3:12

Receiver: the value a method is attached to, named before the method.

Steps0:54

1. go mod init · 2. write main.go · 3. go run .

Compare

Buffered: send blocks only when full. Unbuffered: send always waits.

Flashcards43:55

Q: When does a buffered send block? A: When the buffer is full.

Quiz47:20

What does select do when no case is ready? → It blocks, unless default.

Flow diagram26:40

main → spawn workers → fan-in results → close

Mind map

concurrency: goroutines · channels · select

Timeline

09:00 setup · 09:12 first goroutine · 09:26 selects

Architecture

client → api → service → repo

Table

kind | cue | tool cards | 43:55 | study.create_flashcards

Sketch

Hand-drawn sketch of a worker pool

Interactive

Run this example inside the note. Sandboxed, yours.

index

learning_path

simulation

learning

Transcript14 cues43:55
Open
How: receivers
Desk: errors are values
Blocks: goroutines
Review: channels
Agents: select
03:12

Three moves, then you're studying

01

Hand an agent the desk

Connect Claude, ChatGPT, Cursor, or another MCP client with one scoped token. It uses the same tools the workspace uses.

02

It writes native blocks

Quizzes, cards, flows, callouts. The agent writes the same NoteScript the editor compiles, not a dump of HTML.

03

You study what landed

Open the notebook. Cards are already in review. Every block still knows the second it came from, and which tool wrote it.

12:04

A desk, not a file tree.

Six cutting-mat pigments, one per subject. Hover the stack and the colour follows its notebook everywhere: tabs, headers, and every card an agent sends to review.

Agents write into the notebook you already openedSub-notes for sections that outgrow a pageWiki-links and backlinks between any two notes
26:40

Write it the way the idea is shaped.

23 native types. Press /, or let the agent pick. Hover the belt to slow it down.

TextHeadingQuoteCalloutFormulaCode runnerImageDefinitionStepsCompareFlashcardsQuizFlow diagramMind mapTimelineArchitectureTableSketchInteractiveindexlearning_pathsimulationlearningTextHeadingQuoteCalloutFormulaCode runnerImageDefinitionStepsCompareFlashcardsQuizFlow diagramMind mapTimelineArchitectureTableSketchInteractiveindexlearning_pathsimulationlearning
learningsimulationlearning_pathindexInteractiveSketchTableArchitectureTimelineMind mapFlow diagramQuizFlashcardsCompareStepsDefinitionImageCode runnerFormulaCalloutQuoteHeadingTextlearningsimulationlearning_pathindexInteractiveSketchTableArchitectureTimelineMind mapFlow diagramQuizFlashcardsCompareStepsDefinitionImageCode runnerFormulaCalloutQuoteHeadingText
00:34

Let the explanation move with the idea.

Turn a concept into a short narrated lesson. An agent writes the scenes in NoteScript; NotesKit times the visuals to the words and keeps captions and transcript close by.

Local narration, or a silent timer when voice is offStage actors connect, move, and carry across scenesWord-synced captions and a transcript you can revisit
Video lesson · NoteScript
41:10

The cards write themselves while the agent takes notes.

Mark a definition, get a card. Four grades, keyboard only, and every card carries the cue and the tool that taught it.

3/4 graded
58:30

Humans and agents write the same notes.

One scoped token. The agent gets the workspace tools (create, append, quiz, flow, revise) and every edit lands in history with the tool that made it. No scraping. Revoke in a click.

Connected · MCP session
notescript.agent_guide() notes.create(notebook: "Systems") notes.append_notescript(type: "flashcards", cue: "43:55") study.create_quiz(limit: 6)

Every agent edit lands in the note's history with the tool that made it.

Free while in beta

Open a notebook. Hand it to your agent.

Start with one desk. Connect Claude, ChatGPT, Cursor, or another MCP client when you want it to write.