JarvisHub: An Open Harness for Canvas-Native Multimodal Creative Agents
JarvisHub frames the creative canvas as both workspace and agent memory.
The paper proposes an open harness for agents handling long-running multimodal creative projects, not single prompt-output jobs. It represents assets, versions, dependencies, feedback, and tool actions as typed nodes and links on an editable canvas. Its architecture has three layers: canvas state, protocol bridge, and agent runtime. The stated goal is to make agent planning, revision, organization, and failure recovery inspectable and steerable by humans throughout the process. HF Daily Papers' note
The paper proposes an open harness for agents handling long-running multimodal creative projects, not single prompt-output jobs. It represents assets, versions, dependencies, feedback, and tool actions as typed nodes and links on an editable canvas. Its architecture has three layers: canvas state, protocol bridge, and agent runtime. The stated goal is to make agent planning, revision, organization, and failure recovery inspectable and steerable by humans throughout the process. HF Daily Papers' note
score 5