Skip to content

CLAUDE.md

Latest commit

 

History

History
127 lines (99 loc) · 3.35 KB

CLAUDE.md

File metadata and controls

127 lines (99 loc) · 3.35 KB
Raw

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

This is "Open AI Search" - a conversation-based search application with LLM support that includes Deep Research capabilities. It's built as a monorepo using Turborepo with two main applications: a Node.js/Koa backend server and a Vue.js frontend.

Architecture

Monorepo Structure

  • Root: Turborepo workspace with yarn@3.5.1 package manager
  • apps/server: Node.js backend (Koa.js framework, TypeScript)
  • apps/web: Vue.js frontend (Vue 3, TypeScript, Vite, TDesign UI)

Key Technologies

  • Backend: Koa.js, TypeScript, LangChain, multiple LLM providers (OpenAI, Google, DeepSeek, etc.)
  • Frontend: Vue 3, TypeScript, Vite, Pinia (state management), TDesign Vue Next, Tailwind CSS
  • Build System: Turborepo for monorepo management
  • Search Engines: SearXNG, Bing, Google, Tavily integration
  • AI Features: Deep Research mode with iterative search and analysis

Development Commands

Root Level (Turborepo)

# Start development servers for both apps
turbo dev
# or
yarn dev

# Build all applications
turbo build
# or
yarn build

Server (apps/server)

# Development with hot reload
yarn dev

# Production start
yarn start

# Type checking
yarn check-types

# Build TypeScript
yarn build

Web (apps/web)

# Development server
yarn dev

# Build for production
yarn build

# Preview production build
yarn preview

Key Features & Components

Deep Research Mode

  • Iterative search and analysis workflow
  • Uses LangChain and LangGraph for orchestration
  • Supports multiple search engines and web scraping
  • Generates comprehensive reports through recursive exploration

LLM Provider Support

The application supports multiple LLM providers configured through model.json:

  • OpenAI (including o1 models)
  • Google Gemini
  • DeepSeek (including R1)
  • Ollama, LMStudio
  • Various Chinese providers (Baidu, ChatGLM, Moonshot, Tencent, etc.)

Search Engine Integration

  • SearXNG (free, no API key required)
  • Bing Search API
  • Google Search (via SearchApi, Serper, or Programmable Search)
  • Tavily (LLM-optimized search)
  • Jina Reader for web content extraction

Configuration

Environment Files

  • Development: apps/server/.env.local and apps/server/.env
  • Docker: deploy/.env.docker
  • Model Configuration: apps/server/src/model.json or deploy/model.json

Required Environment Variables

At minimum, set one LLM provider key:

OPENAI_KEY=your_key
# or other provider keys

For search engines:

SEARXNG_HOSTNAME=http://localhost:8080  # if using SearXNG
BING_SEARCH_KEY=your_key               # if using Bing
# etc.

Development Notes

Server Architecture

  • Main entry: apps/server/src/app.ts
  • Controllers: apps/server/src/controller.ts
  • Middleware: apps/server/src/middleware.ts
  • Interfaces: apps/server/src/interface.ts
  • Serves static files from built web app

Frontend Architecture

  • Vue 3 with Composition API
  • Pinia for state management with persistence
  • TDesign Vue Next for UI components
  • Vue Router for navigation
  • i18n support for internationalization
  • Dark/light mode support

Docker Deployment

The project includes Docker Compose setup in the deploy/ directory with SearXNG integration for search functionality.