openxflow

<!-- markdownlint-disable MD001 MD033 MD041 --> <div align="center"> # Tool List ### Making AI Workflows Simpler, More Open, and More Intelligent A new generation of open-source AI workflow platform based on Skills, MCP, Agent and visual orchestration. [English](./README_EN.md) · [简体中文](./README.md) [](./LICENSE) [](https://github.com/lien0219/xiangflow-ai/stargazers) [](https://github.com/lien0219/xiangflow-ai/forks) [](https://github.com/lien0219/xiangflow-ai/issues) [](https://github.com/lien0219/xiangflow-ai/pulls)       [Quick Start](#-quick-start) · [Core Capabilities](#-core-capabilities) · [Development Documentation](#-documentation) · [Contribute](#-contribute) · [GitHub Issues](https://github.com/lien0219/xiangflow-ai/issues) </div> ## Product Introduction XiangFlow AI is an open-source AI workflow platform for developers and enterprises. Through a visual canvas, developers can connect large language models, Agents, knowledge bases, data sources, APIs, and external tools to quickly build, debug, run, and integrate AI applications. The platform combines visual workflows, Agent orchestration, MCP tool protocols, Skills capability organization, and RAG knowledge retrieval, allowing complex AI applications to be built in a modular, reusable, and scalable way. > Let every AI capability be connected, orchestrated, and reused. ## ✨ Product Highlights | Capability | Description | | --- | --- | | Visual Workflow | Connect models, Prompts, tools, knowledge bases, and business logic through node canvas | | AI Agent | Build intelligent agents that understand tasks, select tools, and complete multi-step work | | Skills Extension | Precipitate domain capabilities through reusable processes, prompts, rules, tools, and components | | MCP Ecosystem | Connect external tools as MCP clients and provide workflows as MCP tools | | RAG and Knowledge Base | Combine document processing, embedding, vector retrieval, and context enhancement capabilities | | Multi-Model Support | Connect mainstream cloud models, local models, and compatible API model services | | Custom Components | Extend nodes, tools, data processing, and business integration using Python | | API Service | Integrate workflows into existing systems through REST API, Webhook, and MCP | | Open-Source Deployment | Support local running, container building, private deployment, and customized integration | ## 🧩 Core Capabilities ### Visual AI Workflow - Use node canvas to drag, connect, and configure workflows - Combine models, Prompts, data, tools, and process control components - Pass text, messages, data, and structured results between nodes - Interactively test and view process execution results through Playground - Import, export, and reuse workflows in JSON format - Run workflows through REST API, Webhook, or MCP ### AI Agent - Build Agents in visual workflows, configure models, instructions, and context - Use built-in components, other Agents, and MCP Server as callable tools - Select tools and execute multi-step processing according to tasks - Access session memory, knowledge retrieval, and structured output components - Debug Agent behavior in Playground or API calls ### Skills Skills are the way XiangFlow AI organizes reusable AI capabilities. Domain knowledge, task steps, prompts, rules, tools, and resources can be encapsulated into clear and combinable capability entries through workflows and custom components, and reused in different AI applications. ### MCP - Connect external MCP Server through MCP Tools component - Support STDIO, Streamable HTTP, and compatible SSE connection methods - Configure and use MCP Server in settings or canvas sidebar - Provide workflows in projects as MCP tools to external clients - Support Cursor, Windsurf, Claude Desktop, and other MCP client connections ### RAG and Knowledge Base - Upload and process documents, perform text extraction, segmentation, and metadata management - Use embedding models to generate vector representations - Complete semantic retrieval through knowledge bases and vector storage - Inject retrieval results into model context and build knowledge question and answer processes - Connect files, databases, webpages, and third-party data sources through components ### Model and Component Ecosystem - Integrate multiple large language models, embedding models, and local model services - Provide vector databases, data processing, input/output, and tool components - Write custom components in Python and add them to the visual canvas - Connect business systems through API Request, Webhook, and third-party integration - Extend and run workflows based on LFX execution engine ## 🔄 Workflow ```mermaid flowchart LR User[User Needs] --> Canvas[Visual Workflow] Canvas --> Agent[AI Agent] Canvas --> Skills[Skills Capability Combination] Canvas --> MCP[MCP Tools] Canvas --> RAG[RAG and Knowledge Base] Agent --> Models[Large Language Models] Skills --> Components[Process and Custom Components] MCP --> Services[External Services] RAG --> Data[(Documents and Vector Data)] Models --> Result[AI Applications and API] Components --> Result Services --> Result Data --> Result ``` Starting from needs, combine Agents, reusable capabilities, MCP tools, and knowledge retrieval in the canvas; workflows can be run in the interface or integrated into business systems through API, Webhook, or MCP. ## 🎯 Application Scenarios | Scenario | Description | | --- | --- | | Enterprise Knowledge Base | Connect enterprise documents, data sources, and models to build intelligent Q&A systems with knowledge retrieval capabilities | | AI Customer Service | Combine Agents, knowledge bases, tool calls, and business APIs to build controllable customer service workflows | | Data Analysis Agent | Let Agents connect databases, APIs, and analysis tools to complete query, analysis, and result generation | | Content Generation | Orchestrate models, Prompts, data processing, and output components to build reusable content production processes | | Developer Tools | Connect code tools, editors, and development processes through MCP and reusable capabilities | | Business Process Automation | Connect enterprise systems, external APIs, and AI models to automatically process multi-step business tasks | ## 🏗️ Technical Architecture ```mermaid flowchart TB subgraph Experience[Interaction Layer] Web[React Visual Canvas] Playground[Playground] Clients[API and MCP Clients] end subgraph Application[Application Service Layer] FastAPI[FastAPI REST API] MCPAPI[MCP Server] Services[Authentication, Project, File, and Knowledge Base Services] end subgraph Runtime[Orchestration and Execution Layer] Graph[Workflow Graph Execution Engine] LFX[LFX Runtime] Agents[Agent and Tool Calls] Components[Built-in and Custom Components] end subgraph Integrations[Integration Layer] Models[Model and Embedding Services] MCPTools[External MCP Server] APIs[External API and Data Sources] Vectors[Vector Storage] end Storage[(Application Database and File Storage)] Web --> FastAPI Playground --> FastAPI Clients --> FastAPI Clients --> MCPAPI FastAPI --> Services FastAPI --> Graph MCPAPI --> Graph Graph --> LFX Graph --> Agents Graph --> Components Agents --> Models Agents --> MCPTools Components --> APIs Components --> Vectors Services --> Storage ``` - **Interaction Layer**: Provide visual canvas, Playground, and API and MCP client entries. - **Application Service Layer**: FastAPI carries workflow, project, file, knowledge base, authentication, and MCP interface. - **Orchestration and Execution Layer**: Workflow graph and LFX runtime are responsible for node scheduling, data passing, Agent and component execution. - **Integration Layer**: Connect model services, MCP Server, business APIs, data sources, and vector storage. ## Technology Stack | Domain | Technology | | --- | --- | | Front-end | React 19, TypeScript, Vite, Tailwind CSS, Zustand, XYFlow | | Back-end | Python 3.10–3.14, FastAPI, SQLModel / SQLAlchemy, Alembic | | Workflow | Langflow graph execution engine, LFX | | Agent | LangChain ecosystem, tool calls, structured output | | Protocol | REST API, Webhook, MCP | | Data Storage | SQLite, optional PostgreSQL, file storage | | Vector Retrieval | Chroma, Qdrant, Weaviate, Pinecone, Milvus, FAISS, and other components | | Deployment | Local source code running, Docker / Podman, Docker Compose, Dev Container | ## 🚀 Quick Start ### Environment Requirements - Python `>=3.10,<3.15` - Node.js `>=20.19.0` (recommended v22.12 LTS) and npm v10.9+ - `uv >=0.4` - GNU Make Windows users are recommended to use WSL or the Dev Container provided in the repository. ### Clone Repository Using SSH: ```bash git clone git@github.com:lien0219/xiangflow-ai.git cd xiangflow-ai ``` Or using HTTPS: ```bash git clone https://github.com/lien0219/xiangflow-ai.git cd xiangflow-ai ``` ### Local Running Install dependencies, build front-end, and start application: ```bash make run_cli ``` Access <http://localhost:7860>. If you need to clean the front-end build cache and restart, run: ```bash make run_clic ``` ### Development Mode ```bash make init ``` Start back-end and front-end in two separate terminals: ```bash make backend ``` ```bash make frontend ``` The back-end listens on `http://localhost:7860` by default, and the front-end development server listens on `http://localhost:3000`. For more environment configurations, testing, and component development instructions, please see [Complete Development Guide](./DEVELOPMENT.md). ### Container Building and Running The repository provides a container build target that can be used to build the current source code using Docker: ```bash make docker_build DOCKER=docker docker run --rm -p 7860:7860 langflow:1.10.2 ``` ## Project Structure ```text xiangflow-ai/ ├── .github/ # GitHub workflows and repository configuration ├── deploy/ # Deployment and observability configuration ├── docker/ # Container building and development configuration ├── docker_example/ # Docker Compose example ├── docs/ # Docusaurus documentation ├── scripts/ # Build, test, and maintenance scripts ├── src/ │ ├── backend/ # FastAPI API and application services │ ├── frontend/ # React / TypeScript web application │ ├── lfx/ # Lightweight workflow executor │ ├── langflow-stepflow/ # Workflow step execution support │ ├── sdk/ # Python SDK │ └── bundles/ # Optional component extension packages ├── README.md # Simplified Chinese product introduction ├── README_EN.md # English product overview ├── DEVELOPMENT.md # Local development, Dev Container, and environment configuration └── CONTRIBUTING.md # Contribution guide ``` ## 📚 Documentation | Document | Description | | --- | --- | | [DEVELOPMENT.md](./DEVELOPMENT.md) | Local development, Dev Container, and environment configuration | | [CUSTOMIZATION.md](./CUSTOMIZATION.md) | Project maintenance and customization development guide | | [CONTRIBUTING.md](./CONTRIBUTING.md) | Contribution guide | | [SECURITY.md](./SECURITY.md) | Security strategy and vulnerability reporting method | | [CODE_OF_CONDUCT.md](./CODE_OF_CONDUCT.md) | Community behavior criteria | | [LICENSE](./LICENSE) | MIT License and copyright statement | ## 🤝 Contribute Welcome community developers to participate in XiangFlow AI. You can submit bugs, propose feature suggestions, improve documentation, develop components, fix issues, or submit Pull Requests. - [Submit Issue](https://github.com/lien0219/xiangflow-ai/issues) - [View Pull Requests](https://github.com/lien0219/xiangflow-ai/pulls) - [Read Contribution Guide](./CONTRIBUTING.md) ## 🙏 Acknowledgements The birth of XiangFlow AI is inseparable from the excellent open-source ecosystem. Special thanks to the [Langflow](https://github.com/langflow-ai/langflow) team and all community contributors for their outstanding contributions to visual AI workflow and Agent development ecology. Thanks to LangChain, FastAPI, React, and all related open-source projects and contributors. ## 📄 License This project is open-sourced under the [MIT License](./LICENSE).