Stardust

Open-source AI Agent Platform — Build, Extend, Automate.

Stardust is a desktop AI agent platform that combines a powerful chat interface with an extensible plugin system. It integrates with 60+ AI models, supports MCP (Model Context Protocol) tools, sandboxed code execution, and multi-project management — all wrapped in a clean, modern UI.

Features

🧠 AI Chat

• Multi-model support — OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Qwen, Llama, and 50+ more
• Streaming responses with token usage tracking
• Skills system for reusable prompt workflows
• Agent system with tool-augmented autonomous execution
• Conversation management with Supabase sync

🔌 Plugin System

• React/TSX plugins — full access to shadcn/ui components and React ecosystem
• 60+ host APIs — AI chat, file system, sandbox execution, Supabase, Cloudinary, HTTP, and more
• AI tool injection — plugins can register custom Function Calling tools for the AI to use
• Permission model — declarative permissions in manifest.json, sandboxed file system access
• Crash isolation — each plugin wrapped in ErrorBoundary, single plugin failure won't crash the app
• Community Plugins — browse and install community plugins at stardust-community-plugins

🛠 Built-in Tools

• MCP Gateway — connect to any MCP-compatible server (SSE, stdio, Streamable HTTP)
• Sandbox — execute JavaScript (with npm packages) and Python (with pip packages) in isolation
• File System — read, write, list, and manage files with workspace integration
• Document Conversion — convert PDF, DOCX, PPTX, XLSX to Markdown
• Web Search — integrated search capabilities
• Agent Orchestration — delegate complex tasks to autonomous agents

🗂 Multi-Project

• Project-based workspace with independent contexts
• Per-project file trees and outputs
• Memory system (local + Supabase) for persistent knowledge

🎨 Modern UI

• Dark/light theme with Tailwind CSS
• shadcn/ui component library
• Responsive sidebar navigation
• Customizable layout per plugin

Tech Stack

Quick Start

Prerequisites

• Node.js 20+
• npm

Install & Run

# Clone the repository
git clone https://github.com/Tianzeyui/stardust.git
cd stardust

# Install dependencies
npm install

# Start development
npm run dev

Build & Package

# Build for current platform
npm run build:electron

# Or target specific platforms
npm run build:mac
npm run build:win
npm run build:linux

Build outputs are placed in the release/ directory.

Plugin Development

Plugins use React/TSX. Start by copying the template from stardust-community-plugins:

git clone https://github.com/Tianzeyui/stardust-community-plugins.git
cp -r stardust-community-plugins/_template plugins/my-feature

Directory Structure

my-feature/
  manifest.json     ← Plugin declaration (id, name, permissions, AI tools)
  index.tsx         ← Entry point — export register(ctx) function
  package.json      ← Dependencies (required, can be empty)
  supabase/         ← Database migrations (optional)

Minimal Plugin

import React from 'react'

export function register(ctx: any) {
  const MyPage = () => <div className="p-4">Hello from my plugin!</div>

  ctx.registerNav({ id: 'my-feature', label: 'My Feature', icon: 'Package', order: 90 })
  ctx.registerRoute('my-feature', () => Promise.resolve({ default: MyPage }))
}

Host APIs Available

Registering AI Tools

Plugins can inject Function Calling tools so AI can interact with plugin features:

ctx.onToolRegister((tools) => {
  tools['my_search'] = {
    description: 'Search my plugin data',
    inputSchema: { type: 'object', properties: { query: { type: 'string' } } },
    execute: async (args) => {
      // Your business logic here
      return `Results for: ${args.query}`
    },
  }
})

Or declaratively in manifest.json:

{
  "tools": [{
    "name": "my_tool",
    "description": "What this tool does",
    "inputSchema": { "type": "object", "properties": { "param": { "type": "string" } } },
    "outputTemplate": "Result: ${param}"
  }]
}

Architecture

┌──────────────────────────────────────────┐
│              Electron Shell               │
│  ┌─────────────┐  ┌───────────────────┐  │
│  │  Main Process│  │  Renderer Process │  │
│  │              │  │                   │  │
│  │  • File I/O  │  │  • React UI       │  │
│  │  • Sandbox   │◄─┤  • Plugin System  │  │
│  │  • esbuild   │  │  • AI SDK         │  │
│  │  • MCP Svr   │  │  • Tools Registry │  │
│  └─────────────┘  └───────────────────┘  │
└──────────────────────────────────────────┘

• Main Process — handles file system, sandbox execution, esbuild plugin compilation, MCP server connections, and HTTP proxying
• Renderer Process — React UI with plugin system, AI chat, tools registry, and Supabase integration
• IPC Bridge — contextBridge exposes secure, permission-filtered APIs to plugins

Contributing

Contributions are welcome! Here's how you can help:

1. Create plugins — build plugins using the community plugins template
2. Add APIs — extend src/lib/pluginTypes.ts and src/lib/pluginSystem.ts to add new host capabilities
3. Improve the core — fix bugs, improve performance, enhance the UI

See the plugin documentation in the community plugins repo under _template/docs/ for detailed development guides.

Release Notes

• v0.22.1 — Agent output refactor, timeline rendering, agent card, GitHub Skill install
• v0.22.0 — Community plugins, repo management, Skill toggles
• v0.21.0 — Project system, toolbox upgrade, UI redesign

License

Apache 2.0 © 沉浸位工作室