Humans-in-the-loop can access in-progress invocations to observe browser-based automations in real-time, resolve errors, or take unscripted actions.

To access Live View, print the browser_live_view_url to the console after you’ve invoked your app:

import { Kernel } from '@onkernel/sdk';
import { chromium } from 'playwright';

const kernel = new Kernel();
const app = kernel.app('browser-automation');

const playwrightScript = async (runtimeContext, payload) => {
	const kernelBrowser = await kernel.browsers.create({ invocation_id: runtime_context.invocation_id });
    // Add this line
    console.log(kernelBrowser.browser_live_view_url);
    
	const browser = await chromium.connectOverCDP(kernelBrowser.cdp_ws_url);

	try {
		const context = await browser.newContext();
		const page = await context.newPage();
        await page.waitForTimeout(15000); // Pause for 15 seconds to visit the live view
		await page.goto("https://www.google.com");
		const title = await page.title();
	} catch (error) {
		console.error(error);
	} finally {
		await browser.close();
	}
};

app.action("my-action-name", playwrightScript);

In a separate terminal, tail the logs to print the live view URL:

kernel logs browser-automation --follow

There currently isn’t a way to end a invocation from the live view.

​
Browser persistence

If you marked a browser for persistence, browser_live_view_url remains active after the invocation ends. It remains active until you terminate the browser.

For browsers not marked for persistence, browser_live_view_url becomes invalid once the invocation’s status reaches a terminal state (SUCCEEDED or FAILED).