DesktopChat

A powerful, local-first desktop application for interacting with and managing AI assistants, agents, and knowledge bases.

Features

• Assistants: Create, manage, and chat with highly configurable AI assistants that can leverage local knowledge bases.
• Agents: Manage and utilize a roster of pre-defined and custom agents for various tasks.
• Knowledge Bases: Build and manage vector-based knowledge stores from local files (PDF, TXT, MD), enabling powerful Retrieval-Augmented Generation (RAG) capabilities.
• File Management: A centralized location for managing all user-uploaded files.
• Extensive Settings: A detailed settings panel for configuring LLM providers, default models, web search APIs, data storage (including Qdrant), and application personalization.

Implemented Components

Frontend (React + TypeScript)

• Main Layout: Sidebar navigation with access to all sections
• Assistants Page: Create and manage AI assistants with chat interface
• Agents Page: Browse and use specialized agents
• Knowledge Page: Create and manage knowledge bases
• Files Page: Upload and manage local files
• Settings Pages:
  • General settings (user profile, language, theme)
  • Provider settings (API keys for LLM providers)
  • Model settings (default models)
  • MCP Servers (Model Configuration Protocol servers)
  • Data settings (Qdrant configuration)
  • Web Search settings (Brave, Google APIs)
  • About page (application information)

Backend (Deno + TypeScript)

• Services:
  • Assistant service (create, list, update, delete assistants)
  • Agent service (list available agents)
  • Chat service (start chat sessions, send messages)
  • File service (upload, list, delete files)
  • Knowledge service (create knowledge bases, add files to knowledge bases)
  • Settings service (manage application settings)
• Core Modules:
  • File processor (parse, chunk, and embed files)
  • RAG pipeline (retrieve relevant context and generate responses)
  • LLM factory (abstract different LLM providers)
• Database:
  • SQLite client (metadata storage)
  • Qdrant client (vector storage)

Technology Stack

• Application Framework: Tauri
• Backend Runtime: Deno
• Vector Database: Qdrant
• Metadata Database: SQLite
• Frontend Framework: React with Vite
• State Management: Zustand & TanStack Query
• Styling: Tailwind CSS

Project Structure

├── src/                    # React Frontend Source
│   ├── api/               # Tauri command abstractions
│   ├── components/        # Shared UI components
│   ├── features/          # Feature-specific components
│   ├── hooks/             # Custom React hooks
│   ├── pages/             # Top-level page components
│   ├── stores/            # Zustand global state
│   └── assets/            # Static assets
├── src-deno/              # Deno Backend Source
│   ├── api/               # Type definitions
│   ├── core/              # Core business logic
│   ├── db/                # Database clients
│   ├── lib/               # Utility libraries
│   ├── services/          # High-level services
│   ├── util/              # Utility functions
│   └── main.ts            # Main entry point
├── src-tauri/             # Tauri Rust Core
└── Design/                # Design documents and reference images

Development

1. Install dependencies:

   npm install

2. Run the development server:

   npm run tauri dev

Building

To build the application:

npm run tauri build