Agents & Tool Use
tau-bench
Tests whether customer-service agents can follow policies and use tools while conversing with simulated users.
201items
49subjects
Original Sierra code and tasks: MIT. HAL traces: public analysis release; no separate data license found in the captured export repository.license
agents_and_tool_usedomain
textmodality
item-level responses released
Saturation status: No
About this data
Original customer-service tool-agent-user interactions and explicitly attributed third-party HAL runs.
201 items · 4,330 recorded attempts. Browse item pages to see all observations. Filters and means apply to the current page.
Loading observation records…