Updates / product

Launch interactive: try a browser by hand before you write the code

Pick a browser, proxy and options, open a start page and drive it yourself in live view. The code next to it launches exactly the same browser, and the API option is start_url.

2026-10-17

Before writing a scraper, you usually want to know whether the site works at all with a given browser and proxy: does it load, does it show a challenge, where does the consent banner sit? Launch interactive answers that in a minute, with no code. You open a real fleet browser on a start page and drive it yourself in live view. When it works, the code next to it launches the same browser with the same options. Over the API, it's one launch option: start_url.

Try it in the dashboard

• Open the code generator in your dashboard. Pick the browser, the proxy and any options, the same way you would for code.
• Under Try it live, enter a start page and click Launch interactive. The browser opens on that page in a live view, with full mouse and keyboard control. Headless sessions get an address bar with back, forward and reload.
• Click through the site the way your scraper would: accept the banner, open a category, reach the page with the data. If the site blocks this browser or this exit, you see it straight away. Change the browser or the proxy and launch again.
• When it works, the code next to the form launches exactly that browser, in Node.js, Python, Java, C# or Go. Copy it and add your clicks, or switch the builder to Expert for device emulation, blocking and page actions like screenshots and text extraction.

The same thing over the API: start_url

• "start_url": "https://…" makes the server open the page itself, so no client has to connect. POST https://<wsUrl host>/admin/session/live with { "session_id": "…", "mode": "control" } returns a single-use live link to drive it.
• Your code can join later: connect to wsUrl as usual, and the page is already in browser.contexts()[0]. Until the server has loaded it, the page reads about:blank, so wait with page.waitForURL('http*://**'). The code builder does that for you whenever start_url is set. The start_url recipe has the program in five languages.
• Headless works on Chromium-based browsers with "live_view": true (a screencast). Firefox, Camoufox and WebKit open the start page headful only (VNC). Otherwise the launch returns 400 start_url_not_viewable.

How long it stays open, and what it costs

• Mouse and keyboard input count as activity, and so does an open viewer, so the session stays up while you watch or drive. With no input and no one watching, it ends after inactivity_timeout: 120 s by default, up to 10 minutes on shared plans, 5 on Community and 24 h on dedicated threads. Sessions started from the dashboard use your plan's maximum.
• Closing the viewer in the dashboard ends the session, unless you launched with keep_alive.
• It's a normal session: threads and minutes while it runs, on every plan including the free Community plan.

Who it is for

• Scrapers debugging a block: see the challenge, the redirect or the consent wall yourself, on the exact browser and exit your code uses, instead of guessing from a screenshot.
• People who don't write code yet: check that a site works with a given browser and country before anyone builds anything.
• AI agents and people working together: an agent can hand a live session to a person for a captcha or a login, and take it back over the same connection (live view).

Read more

• Case 58: a person fills part of a form in live view, the code connects and finishes it, in five languages. It also measures that an open viewer keeps a quiet session alive past its inactivity_timeout.
• Launch interactive in the docs and the /launch-interactive overview.
• Live links and viewers: Live view. Watching a session afterwards instead: Recordings.