Python in the browser

import { Aside } from ‘@astrojs/starlight/components’;

Run real Python inside the page via Pyodide, loaded on demand. Access via sdk.python. It runs any Python — numpy, pandas, your own logic — and reaches the outside world only through JS functions you explicitly inject.

Methods

run<T>(code, options?): Promise<{ result: T; stdout: string; stderr: string }>
ready(options?): Promise<void>   // preload the runtime
reset(): void                    // drop the cached runtime
readonly loaded: boolean

interface PythonRunOptions {
  globals?: Record<string, unknown>;  // JS values/functions exposed to Python
  packages?: string[];                // e.g. ['numpy', 'pandas']
  install?: string[];                 // pure-Python wheels via micropip
  onStdout?: (line: string) => void;  // streamed print() output
  onStderr?: (line: string) => void;
}

Compute

const { result } = await sdk.python.run(`
import numpy as np
float(np.mean([1, 2, 3, 4]))
`, { packages: ['numpy'] });

console.log(result); // 2.5

Inject your own callbacks

Python can await async JS functions you pass in globals:

await sdk.python.run(`
for wk in weeks:
    await run_sql(TEMPLATE.replace("{W}", wk))
    log(f"{wk} done")
`, {
  globals: { weeks, TEMPLATE, run_sql: async (s) => fetchData(s), log: console.log },
});

The runtime is a lazy singleton — it loads on the first `run()` (a few seconds) and is reused. Call `sdk.python.ready()` on page open to hide that latency, then `run()` is warm. Injected globals are cleared after each call and don't leak.

No manifest needed — sdk.python stores no artifact data. Keep long jobs chunked and stream progress via onStdout, since heavy compute blocks the main thread.