AgentTrove

AgentTrove is the largest open-source collection of agentic interaction traces to date, released by the OpenThoughts-Agent team. It contains 1,696,847 rows drawn from 219 source datasets spanning code repair, shell scripting, mathematical problem-solving, competitive programming, and general computer-use tasks.

At 1.7 million rows, AgentTrove is 4× the size of the Nemotron Terminal Corpus (430 K rows), the previous largest open-source agentic trace dataset.

---

Data Format

Consistent with the Nemotron Terminal Corpus, all traces in AgentTrove are released in the terminus-2 harness format — a ShareGPT-style conversation layout where each row represents a complete agent trajectory, including tool calls, environment responses, and final reasoning.

All traces were generated using the open-source Harbor agentic evaluation and data-generation framework.

---

Schema

Each row contains the following columns. Note that source datasets have heterogeneous schemas; columns not present in a given source are filled with null.

---

Source Datasets

AgentTrove was assembled by concatenating training datasets generated throughought the course of the OpenThoughts-Agent project. The full list is provided below.

---

Citation

If you use AgentTrove in your research, please cite:

@misc{openthoughts-agent,
  author = {Team, OpenThoughts-Agent},
  month = Dec,
  title = {{OpenThoughts-Agent}},
  howpublished = {https://www.open-thoughts.ai/blog/agent},
  year = {2025}
}