OpenVLHarness combines tool use with persistent multimodal memory for visual agents

A member of the UCLA × Apple team describes OpenVLHarness as a multimodal agent with consistent interfaces for tools such as perception and search, persistent memory for images and other artifacts, and a coding agent that combines evidence across tools. The team reports average gains of 9.7–12.6 points over matched no-tool baselines across 22 benchmarks and eight backbones.