Skip to content

Dataset · open-thoughts

Agent interaction trajectories

open-thoughts/AgentTrove

1.7M open agent trajectories from 219 datasets: code, shell, math, and computer use.

What's inside

The AgentTrove collection contains 1,696,847 rows from 219 source datasets. Trajectories cover code repair, shell scripts, math problem solving, competitive programming, and general computer-use tasks. Language is English, size is 1 to 10 million entries, modality is text.

What you can do with it

  • Train an agentic LLM
  • Perform LoRA fine-tuning for tool-use
  • Evaluate an agent on terminal and code tasks
Tasks
Text generation
Languages
English
Size
1M–10M rows
Downloads this month
9.1K
Likes
207

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("open-thoughts/AgentTrove")