Mahal Systems

Overview › Benchmark catalog

WebLINX

Fits chat-driven browser copilots where a human gives instructions turn by turn. Borrow the demonstration format of paired user utterances and browser actions.

What it asks the agent to do

  • predict the next browser action in a recorded conversation with a user
  • reply to the user in chat while navigating a real site

What to borrow from it

case templateuser simulator

Caveats

Offline replay only, no live execution; size is about 2,300 demos (100K steps); data license differs from the Apache code license and recordings may contain personal details.

Details

Workloadsweb browsing, general assistant
Interactionmulti turn scripted
Environmentstatic
End state checkedNo
Toolsrecorded page snapshots, screenshots
Gradingexact match, programmatic check
Size2,300
Datareal scraped
Personal datapossible
LicenseApache-2.0
Harnessown repo
Statusactive

Links