gijl/AgentTrove

AgentTrove是迄今为止最大的开源智能体交互轨迹集合，由OpenThoughts-Agent团队发布。它包含来自219个源数据集的1,696,847行数据，涵盖代码修复、Shell脚本编写、数学问题求解、竞争性编程和通用计算机使用任务。拥有170万行数据，AgentTrove的规模是先前最大的开源智能体轨迹数据集Nemotron Terminal Corpus（43万行）的4倍。AgentTrove is the largest open-source agent interaction trajectory collection to date, released by the OpenThoughts-Agent team. It contains 1,696,847 rows of data from 219 source datasets, covering code repair, shell script writing, mathematical problem-solving, competitive programming, and general computer usage tasks. Boasting 1.7 million rows of data, AgentTrove is four times the size of the previous largest open-source agent trajectory dataset, the Nemotron Terminal Corpus (430,000 rows).