An open-source, full-stack AI Agent & Knowledge Base Platform. It empowers teams to secure their data while leveraging cutting-edge AI capabilities through a private, customizable environment.
graph TD
User["User"] -->|Interacts with| Assistant["Assistant"]
subgraph "RiceBall Platform"
Assistant -->|Based on| Model["Model"]
Assistant -->|Retrieves| RAG["Knowledge Base (RAG / ChromaDB)"]
Assistant -->|Calls| Tools["Tools"]
Model -->|Configured from| Provider["Model Provider"]
Tools -.->|Standard Protocol| MCP["MCP (Model Context Protocol)"]
Scheduler["Scheduler (Cron)"] -->|Triggers| Assistant
end
ExternalApps["External Apps<br/>(WeCom/Telegram)"] -->|Webhook| Assistant
Assistant -.->|Push Message| ExternalApps
Provider -.->|API| ExternalServices["External AI Services (OpenAI/Ollama/etc.)"]
- 🔐 Private RAG Knowledge Base:
- Upload documents (PDF, DOCX, XLSX, PPTX, Markdown) securely.
- Web Reader: Extract clean content from any webpage URL to your knowledge base.
- Automatic chunking and vectorization using ChromaDB.
- Citations and reference tracking.
- Memory & Write-Back: Empower Agents to save conversation summaries, notes, or new knowledge directly back into your Knowledge Base.
- 🐍 Code Interpreter (Sandbox):
- Secure Execution: Run authorized Python code in a secure, isolated Docker container (Sandbox).
- Data Analysis: Perform complex calculations, data analysis, and visualization using Pandas, NumPy, Matplotlib, etc.
- File Processing: Generate and handle files (CSV, Excel, PDF, Images) seamlessly within the conversation.
- 🤖 Agent Engine:
- Built on LangChain, utilizing robust Tool Calling capabilities to execute tasks.
- Universal MCP Support: Full implementation of Model Context Protocol (MCP).
- Stdio: Connect to local tools via standard input/output.
- HTTP (SSE): Uses HTTP POST for client-to-server messages with optional Server-Sent Events, enabling remote or Dockerized MCP server connections.
- Presets: One-click install for common tools like Filesystem, Brave Search, etc.
- 📢 Multi-Channel Integrations:
- Omnichannel: Deploy your Assistant to WeCom (App / Group Robot), Telegram, and more.
- Guest Management: Automatically maps external user identities to internal "Guest" accounts with isolated conversation history.
- Streaming: Supports streaming responses even on platforms that require polling (like WeCom Smart Bot) via an intermediate stream buffer.
- ⏰ Scheduled Tasks:
- Cron Automation: Automate AI interactions with Cron-based scheduling.
- Push Notifications: Send proactive messages to dedicated channels (WeCom, Telegram).
- Targeted Delivery: Support sending to generic targets (Group Chats, Departments) or specific users.
- 🔌 Multi-Model Support:
- Vendor Agnostic: Support OpenAI, Anthropic, Google Gemini, XAI (Grok), etc.
- Ollama Integration: One-click scan & import for local LLMs (DeepSeek, Llama3).
- Cost Optimization: Route simple queries to cheaper models and complex reasoning to high-performance models.
- 🛡️ Enterprise SSO & Auth:
- Full OAuth 2.0 support (Google, GitHub, Keycloak, Auth0).
- Native integration with WeCom (Enterprise WeChat) for seamless employee login.
- 🚀 OpenAI Compatible API:
- Expose your configured Assistant as a standard OpenAI API endpoint.
- Integrate easily with existing tools like VS Code extensions or third-party wrappers.
- Frontend: Nuxt 4 (Vue 3), Shadcn Vue, TailwindCSS v4, Pinia
- Backend: FastAPI, SQLAlchemy (Async), Alembic
- AI & Data: LangChain, ChromaDB, Pydantic
- Storage: PostgreSQL / SQLite, Redis, S3 (MinIO/AWS)
- Security First: Self-hosted solution ensures your IP and user data remain on your infrastructure.
- Business Native: Through Tool Calling and Universal MCP Support, RiceBall isn't just a chatbot—it's an operator that can interact with your business systems (Filesystem, Databases, APIs) regardless of where they are hosted.
- Developer Ready: Clean, modular architecture (Frontend/Backend separation) makes it an excellent starter kit for custom AI solutions.
- Enterprises: Private knowledge assistants and internal tooling.
- Developers: A robust, modern foundation for RAG/Agent applications.
- Integrators: Deploy tailored AI solutions for clients.
The name RiceBall comes from a hamster I own. Every time it eats in its little food bowl, it curls up like a rice ball, so I named it RiceBall.
Note: This setup is for preview and testing purposes only. Please use with caution in production environments.
Run RiceBall with a single command using our All-in-One Docker image (includes SQLite & Local Storage):
docker run -d \
-p 8000:8000 \
-e SUPERUSER_EMAIL=admin@admin.com \
-e SUPERUSER_PASSWORD=admin123456 \
-v riceball_storage:/app/storage \
--name riceball \
ghcr.io/riceball-ai/riceball:all-in-one-latestVisit http://localhost:8000 to start using RiceBall. You can configure the initial superuser credentials via the SUPERUSER_EMAIL and SUPERUSER_PASSWORD environment variables.
For production environments requiring PostgreSQL and S3, you can deploy from source:
git clone https://github.com/riceball-ai/riceball.git
cd riceball
docker compose -f docker-compose.prod.yml up -d
RiceBall wouldn't exist without the contributions of the open-source community. Special thanks to the following excellent open-source projects:
It is impossible to list all projects. If your project is used but not listed here, please contact us to add it.
This project is open source under the MIT License.