PaiAgent: Enterprise AI Workflow Orchestration Platform

PaiAgent is an enterprise-grade AI workflow orchestration platform designed for visually building, composing, and executing complex AI processes. It targets developers and business users seeking to simplify the integration and automation of multiple large language models and tools, offering a no-code, drag-and-drop interface for rapid AI solution development.

How It Works

The platform employs a dual-engine architecture, featuring a self-developed DAG engine for topological sorting and cycle detection, alongside the LangGraph4j state graph engine for advanced conditional branching and dynamic routing. AI model integration is unified through Spring AI, supporting providers like OpenAI, DeepSeek, and Qwen. A novel Skills system allows declarative definition of AI capabilities via YAML and Markdown, with progressive loading to optimize token usage and integration with LLM function calling. Extensibility is facilitated through a plugin architecture for custom nodes, and real-time debugging is supported via SSE streaming.

Quick Start & Requirements

• Prerequisites: Java 21+, Node.js 18+, MySQL 8.0+, Maven 3.8+.
• Setup: Clone the repository, configure backend (.env) and frontend (.env.local) environment variables, start the backend (./mvnw spring-boot:run), and then the frontend (npm install, npm run dev).
• Default Credentials: admin / admin123 (configurable via environment variables).
• Documentation: USER_GUIDE.md, PROGRESS.md.

Highlighted Details

• Visual workflow editor powered by ReactFlow.
• Unified LLM node support for OpenAI, DeepSeek, Qwen, Zhipu AI, and AIPing via Spring AI.
• Tool nodes include TTS, Input/Output, and custom extensions.
• Dual execution engines: DAG and LangGraph4j, selectable via engineType.
• Skills system with declarative definition, progressive loading, and LLM function callback integration.
• Real-time debugging panel with SSE stream output.

Maintenance & Community

The project is actively maintained by @itwanger and welcomes contributions via GitHub Issues and Discussions.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive MIT license suitable for commercial use and integration into closed-source projects.

Limitations & Caveats

The LangGraph4j engine is currently in beta. Key features like conditional branching, loops, user registration, and password management are planned for future releases (v1.1-v2.0). Default development credentials require secure replacement in production environments.