@QingQ77: Added multi-workspace isolation and multi-role permission management on top of DeerFlow, achieving workspace-level MCP tool and skill isolation. https://github.com/gaoyakunsan/deer-flow… Each workspace has its own independent chat history, memories, MCP tools…

On top of DeerFlow, we added multi-workspace isolation and multi-role permission management, enabling space-level MCP tool and skill isolation. https://t.co/P540ygQFxX Each workspace has its own independent chat history, memory, MCP tools, and skills. Three roles—admin, space admin, and member—each manage their own scope. Backend: FastAPI + LangGraph, Frontend: Next.js 16, runs with Docker Compose. https://t.co/wk7ZKIMgsx — # gaoyakunsan/deer-flow Source: https://github.com/gaoyakunsan/deer-flow # 🦌 DeerFlow 2.0 Open Source AI Multi-Space Super Agent Platform Based on DeerFlow (https://github.com/bytedance/DeerFlow) with secondary development, adding multi-space, multi-role permission management, and space-level tool and skill isolation. Python 3.12+ (https://www.python.org/downloads/) Node.js 22+ (https://nodejs.org/) License — ## ✨ Key Highlights Compared to the upstream DeerFlow, this project adds the following core capabilities: | Feature | Description | |———|———––| | 🏢 Multi-Space Isolation | Supports creating multiple independent workspaces; conversation history, agent memory, and files are fully isolated | | 👥 Multi-Role Permissions | Each space supports Admin and Member roles; admins can manage members and space configuration | | 🔧 Space-Level MCP Tools | Each space can independently configure MCP servers; different spaces have different tool sets | | ⭐ Space-Level Custom Skills | Custom skills are isolated per space; each space can independently create, import, and manage skills | | 🔄 Smart Fallback | When a space has no dedicated MCP configuration, it automatically falls back to the global extensions_config.json | | 🛡️ System Admin | New system admin role that can manage all spaces and their members | ## 📸 Interface Screenshots Smart conversation homepage| Multi-space switching| MCP tool management| Custom skill management — ## 🏗️ Architecture Overview ┌─────────────────────────────────────────────────────┐ │ Nginx (2026) │ │ Unified Reverse Proxy Entry │ ├──────────────┬──────────────────────────────────────┤ │ Frontend │ Gateway API │ │ Next.js 16 │ FastAPI (8001) │ │ Port 3000 │ │ │ │ ┌──────────────────────────────────┐│ │ │ │ SpaceMiddleware ││ │ │ │ X-Space-Id → Member Validation → Context ││ │ │ └──────────────────────────────────┘│ │ │ │ │ │ ┌──────────────────────────────────┐│ │ │ │ Agent Runtime (LangGraph) ││ │ │ │ 18 Layers Middleware → Tool Assembly → Sub‑Agent ││ │ │ └──────────────────────────────────┘│ │ │ │ │ │ ┌──────────────────────────────────┐│ │ │ │ Multi‑Space Data Layer ││ │ │ │ Spaces → Members → MCP → Skills ││ │ │ └──────────────────────────────────┘│ └──────────────┴──────────────────────────────────────┘ ### Tech Stack - Backend: Python 3.12+ / FastAPI / LangGraph / SQLAlchemy - Frontend: Next.js 16 / React / Tailwind CSS / TanStack Query - Sandbox: Docker DooD / K8s Provisioner - Deployment: Docker Compose / Nginx ## 🚀 Quick Start ### Prerequisites - Python 3.12+ - Node.js 22+ & pnpm 10.26.2+ - Docker (for sandbox features) - Windows users: Use Git Bash to run all commands ### Installation bash # Clone the repository git clone https://github.com/your-org/deer-flow.git cd deer-flow # Interactive setup wizard (generates config.yaml + .env) make setup # Install all dependencies make install ### Start bash # Development mode (hot reload) make dev # Or Docker mode make docker-init && make docker-start Visit http://localhost:2026. On first startup, you will be redirected to /setup to create the admin account. ### Common Commands bash make dev # Start all services in development mode make start # Start in production mode make stop # Stop all services make doctor # Check configuration and system requirements make lint # Code quality check make test # Run tests ## 📖 Multi-Space Feature Details ### Space Management Each space is a fully isolated work environment with its own: - Conversation History — chat threads isolated per space - Agent Memory — independent memory store per space - MCP Tools — space-level MCP server configuration - Custom Skills — create and manage skills per space ### Roles and Permissions | Role | Permissions | |——|———––| | System Admin | Manage all spaces and users, system-level configuration | | Space Admin | Manage space members, space-level tools and skills | | Space Member | Use conversations, tools, and skills within the space | ### Space-Level MCP Tools Each space can independently configure MCP servers, supporting three transport methods: - stdio — local process communication - sse — Server-Sent Events - http — HTTP long connection When a space has no enabled MCP server configured, it falls back to the global extensions_config.json configuration. ### Space-Level Custom Skills - Isolated skill creation, import, and management per space - Supports skill version management and edit locking - Skill names are unique within a space ## 🔧 Configuration ### config.yaml Main configuration file supporting models, tools, sandbox, memory, sub‑agent, etc.: yaml # Model configuration models: - name: glm-4.7 provider: openai_compatible ... # Sandbox configuration sandbox: use: deerflow.sandbox.local:LocalSandboxProvider ### extensions_config.json Global MCP server and Skills enable/disable configuration. Falls back to this file when a space has no dedicated MCP configuration. ### Environment Variables | Variable | Description | Default | |–––––|———––|———| | DEER_FLOW_PROJECT_ROOT | Project root directory | Parent directory of backend/ | | DEER_FLOW_CONFIG_PATH | Path to config.yaml | Project root directory | | DEER_FLOW_HOME | Runtime state directory | .deer-flow/ | ## 📂 Project Structure deer-flow/ ├── backend/ # Backend service │ ├── app/gateway/ # FastAPI application │ │ ├── auth/ # Space, MCP Provider │ │ ├── routers/ # API routes (spaces, space_mcp, custom_skills) │ │ └── space_middleware.py # Space request context middleware │ └── packages/harness/ │ └── deerflow/ # Agent framework core │ ├── mcp/ # MCP client & space cache │ ├── persistence/ # Data layer (space, space_mcp, custom_skill) │ ├── runtime/ # Runtime (space context ContextVar) │ └── skills/ # Skill lifecycle management ├── frontend/ # Frontend Next.js application │ └── src/ │ ├── core/space/ # Space API, types, Provider │ ├── core/space-mcp/ # Space MCP management hooks │ └── components/ │ ├── workspace/ # Workspace components (space switching, member management) │ └── admin/ # Admin panel (space management pages) ├── config.yaml # Main configuration file ├── extensions_config.json # Global MCP/Skills configuration └── Makefile # Full‑stack operation commands ## 🤝 Acknowledgements - DeerFlow (https://github.com/bytedance/DeerFlow) — ByteDance’s open-source AI super agent base - LangGraph (https://github.com/langchain-ai/langgraph) — LangChain agent orchestration framework - FastAPI (https://fastapi.tiangolo.com/) — High-performance Python web framework - Next.js (https://nextjs.org/) — React full-stack framework ## 📄 License This project is open-sourced under the Apache 2.0 license.