Run code in 25 languages, right inside a page.

Code that lives in your pages

Write a fenced code block with a language, as you would on GitHub or in Obsidian. In the reading or split view it becomes a cell with a Run button. The code stays in the page’s Markdown, so pages stay portable; what a cell prints, the tables it returns and the charts it draws are saved with the page and come back when you reopen it.

```python
import pandas as pd
runs = pd.DataFrame({"month": ["Jan", "Feb", "Mar"], "km": [42, 51, 63]})
runs.describe()
```

25 languages, on your device

JavaScript, TypeScript, Python, SQL, Lua, Shell (emulated), Ruby, PHP, R, Perl, Go, Rust, C, C++, OCaml, Haskell, Zig, C#, F#, Java, Clojure, Elixir, Erlang, Scheme, Prolog.

Each language downloads the first time you run it (from a few hundred kilobytes for Lua or JavaScript to a few dozen megabytes for Rust), then works offline. Settings → Code runtimes downloads them ahead of time, for planes and trains.

Code runs in a sandboxed frame with one worker per page and language. It can’t read the app, your other pages or the network: cells have no network access unless you allow it for that page.

Work like a notebook

• Run all, run the cells above or below, restart the runtimes and run everything again, clear outputs.
• Insert, move and delete cells from the cell menu, each one undoable step. Outputs follow their cell when it moves.
• Stop interrupts a running cell and keeps its variables; if the code doesn’t stop within 3 seconds, the runtime is restarted. Every run has a time limit.
• Variables lists every value your code has made, with its type and a summary.
• Explain and Fix: the optional on-device assistant explains a cell or suggests a fix you accept or reject as a diff.

Share data between languages

Cells in one page can pass data to each other across languages: share(name, value) in one, use(name) in another. Tables arrive as tables: share rows from JavaScript, then query them in SQL.

```js
share("runs", [{ month: "Jan", n: 4 }, { month: "Feb", n: 7 }])
```

```sql
select sum(n) as total from runs;
```

Input, files and packages

• Cells can ask for input (input(), prompt(), readline) and wait for your answer in the page.
• Files attached to the page are there by name: pd.read_csv("sales.csv"). Files a cell writes can be saved to the page or downloaded.
• Python comes with NumPy, pandas, SciPy, scikit-learn, matplotlib, Altair, DuckDB and more, offline; %pip install adds others once you allow the network for that page.

Run your first code cell in Vellum

1. Open a page

   Create a page, or open any page you already have.

2. Write a code block

   Type three backticks and a language name, for example ```python, then your code, then three backticks to close it.

3. Run it

   Switch to the reading or split view and press Run on the cell (or Shift+Enter in the cell editor). The language downloads the first time.

4. Keep the result

   The output is saved with the page. Open the cell menu for Run all, Variables and the other notebook commands.

In the app

• Run code in a page Python, JavaScript, R, SQL and 20+ languages, on this device.
• Variables See every value your code has made.
• Share between languages share() a table in JavaScript, query it in SQL.
• Input, files and charts Cells ask questions, read your files and draw charts.
• Network and packages Cells reach the web or install packages only if you allow.
• Work like a notebook Run all, run above or below, move and restart cells.
• Explain and Fix The on-device assistant explains code and suggests fixes.
• Languages offline Download languages ahead, for planes and trains.

Questions