Published
Updated
Lightpanda's hostel script counts browser memory, and the Jev clock stays in the post
This topic was created 31 days ago, and the information it contains may have evolved or changed since then.
Lightpanda posted that 129 of its browser agents fit on one laptop against 12 Chrome agents, with 11 times less memory on a booking site, and that a hostel task took 6.7 seconds with Jev against 14.1 seconds with gemini-3.8-flash. The linked script measures resident memory and does not call a model. Its notes for the default San Francisco hostel page are 161 MB against Chrome at 1,654 MB.
Lightpanda posted on September 25 that Jev decides and Lightpanda clicks. The task in the post is browsing San Francisco hostels with parking on a startup budget. The three figures in the post are 129 agents on one laptop against 12 with Chrome, 11 times less memory than Chrome on a real booking site, and 6.7 seconds per task with Jev against 14.1 seconds with gemini-3.8-flash.
The reply links a gist, agents_per_laptop.py, created the same day. The file says an agent’s memory is set by the browser under it, so the script does not load a model and does not take an API key. It opens N browsers on one page, holds them, and records peak resident set size across the process tree. Free memory divided by that per-agent figure is the count of agents the machine holds.
The comment gives five pages on an 8-core laptop with about 20 GB free. A Wikipedia article is 62 MB against Chrome at 1,511 MB. travelodge.co.uk is 113 MB against 1,688 MB. The default URL, a Hostelworld page for San Francisco, is 161 MB against 1,654 MB, which the comment marks as 10 times. agoda.com is 226 MB against 1,683 MB. An Airbnb search is 369 MB against 1,951 MB. Chrome stays in a band a bit under 2 GB. Lightpanda’s number moves with the page.
The same comment says a bot challenge makes the ratio meaningless. booking.com and skyscanner.net served a challenge to both engines on every run the author is describing, and kayak.com did so on and off. Against booking.com the script would print 36 times. The author calls that fictional. The run reads body text and warns when the page does not look like the site.
The gist does not print 129, 12, or 11 times. Twenty gigabytes divided by 161 MB is about 124 agents, and divided by 1,654 MB is about 12. 1,654 divided by 161 is about 10.3. The post’s three memory figures sit next to that arithmetic. The 6.7 second and 14.1 second lines appear only in the post. We did not run the script, and we did not time a hostel task.
This site's reading
Editorial notes evaluating claims against primary sources, contextualizing findings alongside related implementations, and defining technical terms.
Verify
The September 25 post, 15:51 UTC, lists three figures: 129 agents on one laptop against 12 with Chrome, 11 times less memory than Chrome on a real booking site, and 6.7 seconds per task with Jev against 14.1 seconds with gemini-3.8-flash.
A reply links gist.github.com/katie-lpd/61a50cfb30d8926f8f1617503aa8fce3, created the same day at 13:19 UTC. The script's opening comment says there is no model and no API key.
The same comment prints an 8-core laptop with about 20 GB free: en.wikipedia.org 62 MB against Chrome 1,511 MB (24 times), travelodge.co.uk 113 against 1,688 (15 times), hostelworld.com San Francisco 161 against 1,654 (10 times, the default URL), agoda.com 226 against 1,683 (7 times), airbnb.com search 369 against 1,951 (5 times).
It says Chrome's figure barely moves and Lightpanda's figure is mostly the page's JavaScript. booking.com and skyscanner.net answer both engines with a bot challenge every time in that note, and kayak.com does so intermittently.
Measured against booking.com, the script would report 36 times, which the comment calls fictional. The 6.7 second and 14.1 second lines are not in the file. We did not run it.
Compare
Browser Use's flights demo and the Stagehand computer-use clip time a model that clicks. This gist times how many browser processes fit in RAM. The 6.7 second Jev line has no script, no task list, and no second run beside it.
The docstring's 10 times on the default hostel page is the figure the file actually prints. The post's 11 times and 129 against 12 are close to 1,654 / 161 and to 20 GB divided by those two footprints, and they are not the same sentence as the file.
Terms
- agents per laptop
- The gist's count of how many browser processes fit in free RAM. It launches N copies, holds them open, and samples peak resident memory. The default page is a Hostelworld listing for San Francisco.
- bot challenge
- A page that asks the client to prove it is human. The script treats a challenge as a failed comparison, because both engines can be measured against a stub. The comment names booking.com as a case that would print a fictional 36 times.