This project offers an innovative approach to AI agent development by transforming abstract workflows into an intuitive, spatialized, and gamified experience. Targeting beginners and users with zero prior experience, it leverages a "Jumping Agent" game metaphor to simplify agent construction, making the process accessible and engaging, particularly on mobile devices like tablets.

How It Works

The platform reimagines traditional flat agent-building workflows into a dynamic, game-like process. A frontend interface, styled after the "Jumping Agent" game, visualizes the agent's operational flow, allowing users to configure nodes and interact via chat. The backend comprises agent_builder for core templates, back_agent for refining user-defined requirements into functional agents, and an orchestrator that manages requests and agent workspaces. A crucial component is the sandbox environment, which provides isolated tool capabilities (file operations, shell access, browser automation) to ensure agent actions do not compromise the host system. This spatial, step-by-step approach significantly lowers the barrier to entry compared to conventional, diagram-heavy interfaces.

Quick Start & Requirements

• Prerequisites: Git, Python 3.11+, Node.js 18+, npm, Docker.
• Installation: Clone the repository, install frontend (npm install) and backend Python dependencies (pip install "fastapi" "uvicorn[standard]" "pydantic" "openai"). A virtual environment is recommended.
• API Key: Configure OPENAI_API_KEY via environment variable or the backend/set_agent_api_key.py script.
• Sandbox: Run docker run --security-opt seccomp=unconfined --rm -it -p 8080:8080 ghcr.io/agent-infra/sandbox:latest.
• Services: Start back_agent (uvicorn api:app), backend/Orchestrator (uvicorn orchestrator:app), and Frontend (npm run server) in separate terminals.
• Access: Navigate to http://localhost:6301 in a browser. iPad/LAN access requires using the host machine's IP address.
• Links: Official quick-start instructions are detailed within the provided README.

Highlighted Details

• Gamified, spatialized interface for AI agent construction.
• Mobile-first (tablet) design for agent building workflow.
• Includes 7 pre-defined Agent Workflow templates.
• Utilizes an isolated sandbox for secure agent execution.

Maintenance & Community

This project is developed independently. Contributions via Issues or Pull Requests are welcomed. Direct contact is available via email (answeryt@qq.com) or WeChat (answerYTAarun).

Licensing & Compatibility

The project's license is not specified in the provided documentation. Consequently, compatibility for commercial use or integration into closed-source projects cannot be determined without further clarification.

Limitations & Caveats

The current version exhibits occasional sandbox operation failures and offers a limited set of 7 Agent Workflow templates. The orchestration and final build processes are noted as not yet fully stable. A command-line interface (CLI) is planned but not yet implemented.