LYL1015/JarvisHub

What it solves

JarvisHub addresses the limitations of traditional creative AI tools. Prompt-to-output tools often hide the intermediate steps and revision history, while chatbot agents struggle to represent spatial layouts and asset dependencies. Node-based workflows are often too rigid, relying on manual pipelines rather than a dynamic project state that an agent can inspect and modify.

How it works

JarvisHub uses an editable canvas as the single source of truth for a project. This canvas acts as a shared workspace for humans and agents, serving as external memory and a protocol-constrained action space. The system is organized into three layers:

• Canvas State: Stores artifacts, layouts, dependencies, and user feedback in a PostgreSQL database.
• Protocol Bridge: Validates mutations and tool actions, ensuring state transitions are explicit and inspectable.
• Agent Runtime: Observes the canvas, plans actions using Skills, Memory, and Tools, and commits the results back to the workspace.

Who it’s for

It is designed for creators and developers building multimodal creative agents for long-horizon tasks, such as narrative media generation, interactive web development, and presentation deck creation.

Highlights

• Canvas-Native State: Keeps all prompts, drafts, and version lineage visible and editable on one canvas.
• Multimodal Orchestration: Supports tools for image, video, audio, code, browser, and presentations.
• Agentic Capabilities: Includes support for subagents to explore subtasks and a memory system for preferences.
• Traceability: Includes a Trace Viewer to inspect and recover agent trajectories (requests, actions, and observations).