bubench-run

To automate browser agent benchmarking, you can orchestrate the full lifecycle of agent runs and subsequent evaluations using background processes. This allows you to execute and monitor agent performance across multiple models without manual oversight.

Non-blocking log monitoring for performance testing is a mechanism that provides real-time tracking of agent progress, success rates, and error recovery. It tracks automated performance metrics while long-running browser interactions execute in the background.

Yes, you need a local browser backend to run automated performance evaluations. The evaluation process requires integration with local browser backends and structured log analysis to execute browser interactions and ensure reproducible evaluation results.

Yes, you can skip already-processed tasks during large-scale benchmark evaluation. The incremental evaluation feature efficiently handles large benchmarks by skipping processed tasks, allowing seamless multi-pass evaluation across diverse datasets.

Yes, uv works with browser agent benchmark experiments as a required dependency. It supports the execution and evaluation of browser agent benchmark experiments across diverse datasets and model configurations.