ProCUA-SFT: Large-Scale Synthetic Trajectories for Desktop Agents

ProCUA-SFT is a large-scale synthetic trajectory dataset for training computer-use agents (CUAs). It was created by NVIDIA and last updated on June 12, 2026. The dataset is designed for supervised fine-tuning of screenshot-based desktop agents that operate graphical environments using mouse, keyboard, and code-like actions.

Use Cases

• Supervised fine-tuning of screenshot-based desktop agents based on the described trajectory data
• Training models to generate mouse and keyboard actions based on visual input from the synthetic desktop trajectories
• Developing agents that can execute code-like actions within a graphical desktop environment as described

Strengths

• Dataset is described as 'large-scale', suggesting substantial size for training
• Accompanies a technical report, indicating formal documentation
• Created by NVIDIA, a recognized organization in AI research

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Description metadata is limited; actual data quality requires manual inspection after download

Provenance

Source

NVIDIA

Collection Method

Synthetic generation, as indicated by the description

Freshness

Last updated 2026-06-12 17:10:54; freshness should be verified