browser · framework · product

Observable (notebooks, Framework, Notebooks 2.0)

Reactive JavaScript notebooks with D3/Plot, now an open HTML file format plus a desktop app.

watchevidence: mixedby Observable, Inc.observablehq.com ↗

Observable (notebooks, Framework, Notebooks 2.0)

  • Maker: Observable, Inc. (Mike Bostock, creator of D3, and team)
  • URL: https://observablehq.com
  • Status (10 October 2026): hosted notebooks since 2018. Observable Framework (static data apps, ISC, 3.7k stars, last release v1.13.4 on 2 March 2026, last push May 2026) looks quieter. Notebooks 2.0 was previewed in late July 2025: an open HTML-based notebook file format, the open source Notebook Kit (ISC, 379 stars, pushed 9 October 2026) and a macOS desktop app (now “Notebook Desktop”). Observable Canvases is the hosted, collaborative, AI-heavy product. Observable Plot (ISC, 5.4k stars) is the charting library.

What it is

A reactive JavaScript notebook: cells are re-run automatically when what they depend on changes, and outputs are live visualisations, usually D3 or Plot. It is the closest thing the web has to a medium for explorable data explanations, and much of the explorable-explanation community wrote in it.

The problem it’s solving

Exploring and explaining data needs code, live output and prose in one place, and the result should stay reactive so readers can change inputs.

Its path / bet

Reactivity plus the web’s full graphics, in a document. Over time the company has split the bet: open files and local tools for developers (Framework, Notebook Kit, Desktop), and a hosted collaborative canvas with AI for business (Canvases).

How it works (concretely)

Cells form a dependency graph; changing an input reruns dependents. Notebooks 2.0 stores a notebook as an HTML dialect and uses vanilla JavaScript (so standard tools like ESLint work), replacing the Observable JS dialect’s custom syntax such as viewof. Desktop appears to be a Tauri app with sandboxed iframes and a CodeMirror editor. Tom MacWright (first hire, holds equity) describes the notebook AI as a light touch and Canvases as AI-heavy. We couldn’t read Observable’s own pages in detail (rate-limited), so the AI specifics are unverified.

Strengths

  • Reactive dataflow is the right model for explorable data, and it’s proven at scale.
  • Best-in-class web charts (D3, Plot).
  • Moving to open files (“file over app”) makes notebooks something an agent can write.

Weaknesses / limits

  • The author still writes code; the reader’s interaction is what the author built.
  • Product direction has shifted several times (notebooks, Framework, Canvases, Notebooks 2.0), which makes it hard to build on.
  • Browser only.

Relation to fictty

Watch / inspire. Observable’s reactivity is the design fictty’s data sources gesture at: a dependency from a source to every node bound to it. Notebooks are durable documents; fictty screens are short-lived. They would only compete if agents started generating Notebook Kit files as their scratch UIs, which is plausible for data work in a browser.

Could fictty adopt it instead of building?

No. Notebook Kit is a browser notebook runtime in JavaScript; fictty is a terminal runtime. One could argue that for data exploration an agent should just write a Notebook Kit file and open it; for a person in a browser that’s often right.

What fictty should take from it

  • Dependency-driven updates. A parameter (input) that sources depend on, rerunning only what depends on it, is Observable’s core idea and the missing piece for reactive teaching screens.
  • An HTML-ish file format was chosen over JSON for human editing. fictty’s JSON is written by agents; if people are to edit saved screens, consider whether a friendlier authoring surface (YAML or the template markup) is needed, and keep the canonical form JSON.

Sources