Refly.AI
Open Source 'Vibe' AI Workspace
Refly is an open-source Vibe AI workspace based on cutting-edge A2A and MCP protocols, integrating world-class AI models. Whether you have a creative idea, a complex requirement, or just a vague concept, Refly helps you turn it into reusable and shareable AI workflows or applications. This enables you to quickly build demo prototypes or deeply explore and refine your ideas with the assistance of AI.
🚀 v0.7.1 Official Release! MCP Store and Backend Skill Invocation 🚀⚡️
Refly Cloud · Self-hosting · Forum · Discord · Twitter · Documentation
🧵 Multi-threaded Conversation System
Imagine being able to conduct multiple independent conversations simultaneously, each with its own context and memory. Refly's multi-threaded conversation system makes this possible. Based on an innovative multi-threaded architecture, it can manage multiple session contexts in parallel, enabling efficient state management and intelligent context switching. This allows for complex agentic workflows, breaking through the limitations of traditional dialogue models.
- Supports multi-threaded conversations, each with independent context and memory
- Innovative multi-threaded architecture for parallel session management
- Efficient state management and intelligent context switching
- Enables complex agentic workflows, surpassing traditional dialogue model limitations
🤖 Multi-Model Integration Framework
We understand that different AI models have their own strengths, so Refly integrates 13+ mainstream large language models, including DeepSeek R1, Claude 3.5 Sonnet, Google Gemini 2.0, OpenAI O3-mini, and more. The system supports hybrid model scheduling and parallel processing, allowing you to flexibly switch between models and enjoy a unified conversation experience. More importantly, multiple models can collaborate to provide you with more comprehensive knowledge support.
- Integrates 13+ mainstream large language models (e.g., DeepSeek R1, Claude 3.5 Sonnet, Google Gemini 2.0, OpenAI O3-mini, etc.)
- Supports hybrid model scheduling and parallel processing
- Flexible switching between models for a unified conversation experience
- Multi-model collaboration for more comprehensive knowledge support
🎨 Multimodal Processing Capability
Refly not only understands text but can also handle various types of files. We support 7+ document formats including PDF, DOCX, RTF, TXT, MD, HTML, EPUB, as well as mainstream image formats such as PNG, JPG, JPEG, BMP, GIF, SVG, and WEBP. With intelligent batch processing, you can select and analyze multiple elements on the canvas at once, greatly improving work efficiency.
- Supports multimodal content processing, including text, documents (PDF, DOCX, RTF, TXT, MD, HTML, EPUB), and mainstream image formats (PNG, JPG, JPEG, BMP, GIF, SVG, WEBP)
- Provides intelligent batch processing for selecting and analyzing multiple elements on the canvas
- Significantly improves efficiency and workflow automation for handling diverse content types
⚡️ AI-Powered Skills System
We have integrated advanced capabilities such as Perplexity AI and Stanford Storm to provide you with powerful AI skills. The system can perform intelligent web-wide search and information aggregation, precise knowledge retrieval based on vector databases, intelligent question rewriting and recommendations, and assist you in generating professional document workflows.
- Integrates advanced AI capabilities like Perplexity AI and Stanford Storm
- Supports intelligent web-wide search and information aggregation
- Precise knowledge retrieval based on vector databases
- Intelligent question rewriting and recommendations
- Assists in generating professional document workflows
🔍 Context Management System
Understanding context is key to AI work. Refly's context management system can accurately build temporary knowledge bases, provide flexible node selection mechanisms, and establish multi-dimensional context associations. Like the Cursor editor, the system can intelligently understand your intent and provide the most relevant contextual support.
- Supports precise construction of temporary knowledge bases
- Flexible node selection mechanism
- Multi-dimensional context association capabilities
- Intelligently understands user intent and provides relevant context support
📚 Knowledge Base Engine
Your knowledge is your wealth. Refly's knowledge base engine supports multi-source heterogeneous data import, and its RAG-based semantic retrieval architecture makes knowledge search simple and efficient. The system can also intelligently build knowledge graphs, creating a personalized knowledge space for you and making knowledge management easy and enjoyable.
- Supports multi-source heterogeneous data import
- RAG-based semantic retrieval architecture for efficient search
- Intelligent knowledge graph construction
- Personalized knowledge space for easy management
✂️ Intelligent Content Collection
In the age of information overload, efficient content collection is crucial. Refly supports one-click content collection from major platforms such as Github, Medium, Wikipedia, and Arxiv, with intelligent parsing and structured processing, automatic categorization and tagging, and deep integration into your knowledge base, making information acquisition unprecedentedly simple.
- One-click content collection from major platforms (e.g., Github, Medium, Wikipedia, Arxiv, etc.)
- Intelligent parsing and structured processing of collected content
- Automatic categorization and tagging, deep integration into the knowledge base
- Improves information acquisition efficiency and streamlines the collection process
📌 Citation System
Good ideas need good citations. Refly's citation system supports flexible multi-source content citation, intelligent context association, one-click citation generation, and provides complete citation traceability. Every point you make is well-supported, and every conclusion is traceable.
- Supports flexible multi-source content citation
- Intelligent context association and one-click citation generation
- Complete citation traceability support
- Ensures every point and conclusion is well-supported
✍️ AI-Enhanced Editor
Writing shouldn't be a solo battle. Refly's AI-enhanced editor offers real-time Markdown rendering, AI-assisted content optimization, and intelligent content analysis, delivering a smooth editing experience similar to Notion. Let AI be your writing partner and create better content together.
- Real-time Markdown rendering for WYSIWYG editing
- AI-assisted content optimization and intelligent analysis
- Smooth editing experience supporting multiple formats
- Notion-like collaboration and writing experience
🎨 Code Generation Engine
From idea to code in an instant. Refly's code generation engine can generate HTML, SVG, Mermaid diagrams, and React applications, intelligently optimize code structure, support component-based architecture, and provide real-time code preview and debugging. Programming becomes more intuitive and efficient.
- Supports generation of various code types (HTML, SVG, Mermaid, React apps, etc.)
- Intelligent code structure optimization for better maintainability
- Supports component-based development
- Real-time code preview and debugging
- Simplifies the transition from idea to code
🌐 Website Visualization Engine
What you see is what you get. Refly's website visualization engine provides interactive web page rendering and preview, supports visualization of complex concepts, dynamic SVG and chart generation, responsive design templates, real-time website prototyping, and integration with modern web frameworks. Building websites becomes as easy as stacking blocks.
- Interactive web page rendering and real-time preview
- Supports visualization of complex concepts
- Dynamic SVG and chart generation
- Provides responsive design templates
- Real-time website prototyping
- Integration with modern web frameworks
We are continuously iterating on Refly, committed to bringing users more innovative and practical features. Below are the main features we are planning and developing:
| Category | Feature Description |
|---|---|
| 🎨 Generation | - Advanced image, audio, and video generation - Cross-modal content transformation tools |
| 💻 Client | - High-performance desktop client (better resource management) - Enhanced offline capabilities |
| 📚 Knowledge Mgmt | - Advanced knowledge organization and visualization tools - Collaborative knowledge base |
| 🔌 Plugin Ecosystem | - MCP-based third-party plugin open standard - Plugin marketplace and developer SDK |
| 🤖 Agents | - Minimally supervised autonomous task agents - Multi-agent collaboration system |
| ⚡️ Workflow | - Visual AI workflow builder and API support for complex processes - Advanced integration with external systems |
| 🔒 Security & Team | - Enhanced security and compliance tools - Advanced team management and analytics |
For the detailed roadmap, please visit: Full Roadmap Documentation
Refly offers multiple usage options to meet the needs of different users and teams:
- Cloud Experience
- No configuration required. Simply visit https://refly.ai/ to try Refly Cloud for free. This version includes all features of the self-hosted edition, with built-in GPT-4o-mini (free) and limited access to GPT-4o and Claude-3.5-Sonnet.
- Self-hosting (Community Edition)
- Quickly deploy Refly locally or on your server by following the Getting Started Guide. For detailed deployment steps, environment variable instructions, and FAQs, please refer to the official documentation.
- Enterprise/Organizations
- For enterprise-level private deployment, customized support, or large-scale collaboration, please email [email protected] to get a tailored solution.
Before installing ReflyAI, please ensure your machine meets the following minimum system requirements:
| Component | Minimum Requirement |
|---|---|
| CPU | 2 cores |
| Memory | 4GB |
Recommended - Use Docker to quickly deploy a fully functional ReflyAI (our team is working hard to update to the latest version):
cd deploy/docker
cp ../../apps/api/.env.example .env # Ensure all required environment variables are set
docker compose up -dAccess URL:
- Docker: http://localhost:5700
Enterprise Deployment - Deploy ReflyAI in a K8s cluster:
cd deploy/kubernetes
kubectl apply -f refly-deployment.yamlAccess URL:
- Kubernetes:
http://${HOST_IP}:30001
Developer Mode - Contribute to project development:
See CONTRIBUTING for more information.
📖 Detailed Deployment Guide: Deployment Tutorial
Star Refly on GitHub to receive instant notifications about new releases.
| Bug Report | Feature Request | Questions/Discussion | ReflyAI Community |
|---|---|---|---|
| Create Bug Report | Submit Feature Request | View GitHub Discussions | Visit ReflyAI Community |
| Something is not working as expected | Ideas for new features or improvements | Discuss and ask questions | Ask, learn, and connect with others |
We welcome developers, testers, technical writers, and all kinds of contributors to join Refly! We encourage all forms of contribution. You can refer to the CONTRIBUTING.md for detailed contribution guidelines, or browse GitHub issues at any time to participate in discussions or submit suggestions.
If you need to report a bug, request a new feature, or provide other feedback, please create a new issue and select the appropriate template so we can better follow up.
If you have any questions, feel free to contact us. For more information and the best place to communicate and learn, visit the ReflyAI Community and grow together with like-minded partners.
|
Idea exchange, feedback, and Q&A |
Bug reports, feature requests, suggestions |
Real-time interaction, tech exchange, project sharing |
Follow updates, product news, official info |
Chinese community, peer exchange, get support |
We would also like to thank the following open-source projects that make ReflyAI possible:
- LangChain - A library for building AI applications.
- ReactFlow - A library for building visual workflows.
- Tiptap - A library for building collaborative editors.
- Ant Design - A UI component library.
- yjs - Provides CRDTs as the foundation for our state management and data synchronization.
- React - A library for building web and native user interfaces.
- NestJS - A framework for building Node.js servers.
- Zustand - Primal and flexible state management for React.
- Vite - Next-generation frontend tooling.
- TailwindCSS - A CSS library for writing beautiful styles.
- Tanstack Query - A library for frontend data fetching and request handling.
- Radix-UI - A library for building accessible React UI components.
- Elasticsearch - A library for building search functionality.
- QDrant - A library for building vector search features.
- Resend - A library for building email sending features.
- Other upstream dependencies.
We are deeply grateful to the community for providing such powerful and easy-to-use libraries, enabling us to focus more on product logic. We hope that in the future, our project can also provide a more user-friendly Vibe Workflow platform for everyone.
To protect your privacy, please avoid posting security-related issues on GitHub. Instead, send your concerns to [email protected], and we will provide you with a more detailed response.
This codebase uses the ReflyAI Open Source License, which is essentially the Apache 2.0 license with some additional restrictions.