JarvisHub: An Open Harness for Canvas-Native Multimodal Creative Agents

Researchers introduced JarvisHub, an open canvas-based framework for multimodal creative agents.

It uses an editable canvas as shared project state, memory, and action space, letting agents plan, generate, revise, organize, and recover across long workflows.

The approach aims to support complex, multi-step creative projects while making agent behavior more inspectable and controllable than closed commercial systems.

The paper offers an open harness for studying and building creative AI agents with external memory and human-steerable automation.