Architecture Overview

RiceBall is not just a tech stack, but a comprehensive AI Agent & Knowledge Base Platform.

Core Concepts

The diagram below shows the relationships between the core entities in the RiceBall system:

Entity Explanation

  • Assistant: The core entity for user interaction. It defines "Who am I" (System Prompt), "What brain do I use" (Model), "What do I remember" (Knowledge Base), and "What can I do" (Tools).
  • Model Provider & Model:
    • Provider: Manages credentials for AI service providers (e.g., OpenAI API Key).
    • Model: Specific model configuration (e.g., GPT-4o). RiceBall abstracts away the differences between providers.
  • RAG (Knowledge Base): Vector storage based on ChromaDB. Allows users to upload documents, and the Assistant automatically retrieves relevant snippets when answering, enabling "open-book exams".
  • Tools & MCP:
    • Tools: Functional capabilities that the Assistant can call.
    • MCP: Uses the Model Context Protocol to standardize the integration of tools and context, making it very easy to extend Assistant capabilities.

Technical Architecture

RiceBall adopts a modern separated frontend and backend architecture.

Backend

  • Framework: FastAPI
  • Database: Async SQLAlchemy + PostgreSQL (or other supported databases)
  • AI Orchestration: LangChain
  • Protocol: Model Context Protocol (MCP)

Frontend

  • Framework: Nuxt 3 (Vue 3)
  • UI Library: Tailwind CSS + Shadcn Vue
  • State Management: Pinia

Directory Structure

  • backend/: Python backend code
  • frontend/: Vue/Nuxt frontend code
  • src/: Backend core logic modules