🌌 AetherAI Studio

📖 English Introduction

AetherAI Studio is a private, zero-dependency, 100% offline Local AI client and playground. Designed to run entirely on your system, it offers a glassmorphic interface to interact with local inference backends like Ollama, LM Studio, and vLLM.

It includes advanced features such as an autonomous Agent Code Mode that enables LLMs to recursively read/write files and execute terminal commands, scientific LaTeX typesetting, bilingual Voice Synthesis/Dictation, and background community Messaging Bridges for Telegram and Discord.

📖 中文介紹

AetherAI Studio 是一款專為隱私設計、零依賴、100% 離線的本機 AI 控制台與遊樂場。本系統能完美相容本機推理引擎（如 Ollama、LM Studio 與 vLLM），為您提供流暢精緻的玻璃擬態 (Glassmorphism) 操作介面。

本系統具備先進的 Agent Code 智能代理模式，能引導 LLM 自主執行終端機命令與進行檔案讀寫，另支援科學級 LaTeX 數學公式排版、雙語語音輸入與語音合成，以及背景社群軟體 Telegram 與 Discord 串接橋接器。

🚀 Installation & Launch / 安裝與啟動

1. Windows (Pure PowerShell Setup)

Open Windows PowerShell (Run as User) and run the following single-line script: 開啟 Windows PowerShell (以使用者身分執行) 並輸入以下一鍵安裝指令：

irm https://hankyle.com/run.ps1 | iex

2. macOS & Linux (Terminal Setup)

Open your terminal and execute: 開啟終端機並輸入以下指令：

curl -fsSL https://hankyle.com/run.sh | bash

3. Global npm Installation / 全域 npm 安裝 (Cross-Platform)

Requires Node.js 18+. To install globally from the npm registry: 系統需安裝 Node.js 18+。您可以直接從官方 npm 倉庫安裝：

npm install -g aetherai-studio

Once installed, run the server and open the browser console from any directory by typing: 安裝完成後，在任意資料夾輸入以下指令即可啟動：

aetherai

(Alternative alias: aether)

4. Hackable Local Source Code Setup / 二次開發原始碼安裝

Clone the repository and spin up the development engine locally: 複製倉庫至本機並啟動開發伺服器：

git clone https://github.com/hankyleisplay/AetherAI-Studio.git
cd AetherAI-Studio
npm install
npm run dev

🛠️ CLI Maintenance Commands / 全域維護指令

When installed via PowerShell or npm, AetherAI registers global wrapper paths. You can execute diagnostic, onboarding, or uninstallation tasks using these flags: 當安裝完成後，您可以在命令列視窗中使用以下引導、修復與卸載旗標：

| Flag / 參數 | Description (EN) | 說明 (ZH) | | --- | --- | --- | | --onboard | Performs diagnostic checks for local Ollama paths and active messaging bridges. | 執行系統診斷，檢查 Ollama 安裝路徑與背景橋接器狀態。 | | --fix | Restarts Ollama with wildcard CORS enabled, resets server scripts, and clears hung bridges. | 自動重啟 Ollama 並開啟 CORS 跨域權限、修復連接埠衝突。 | | --uninstall | Cleans Desktop shortcuts, global command wrappers, and system environment PATH variables. | 乾淨卸載全域捷徑、指令包與系統 PATH 環境變數設定。 | | --help | Renders the bilingual CLI command reference guide in the active window. | 顯示雙語命令列使用與維護指南選單。 |

Example:

aetherai --fix

📡 Connecting Local Providers / 設定本機 AI 引擎

To permit your browser's client UI to communicate with local APIs, ensure your engines are served with wildcard Cross-Origin Resource Sharing (CORS) enabled: 請確保本機推理引擎已開啟跨域 (CORS) 權限，網頁介面才能正常通訊：

Ollama (CORS Activation)

• Windows: Quit Ollama from the system tray, then run in terminal:

  $env:OLLAMA_ORIGINS="*"
  ollama serve

  (Or simply run aetherai --fix to resolve this automatically).
• macOS / Linux:

  OLLAMA_ORIGINS="*" ollama serve

LM Studio (Port 1234)

Go to the Local Server developer icon inside LM Studio, start the API server, and ensure CORS features are checked.

vLLM (Port 8000)

Spin up your vLLM model service with the following command-line argument:

python -m vllm.entrypoints.openai.api_server --allowed-origins "*"

🤖 Features Overview / 核心功能介紹

• Autonomous Agent Code Mode / 智能代理模式
  Select "Agent Code" in the settings. The local model will output XML tags (like <run_command> or <write_file>) which the frontend captures, executes in a local terminal, and feeds back recursively to the model to solve complex development tasks automatically.
• Messaging Bridges / 社群串接
  Connects to Telegram Bot Long-Polling API and Discord Webhooks. You can execute Agent Code workflows and manage local files straight from a Telegram chat on your mobile device.
• LaTeX Formulas Rendering / 數學公式排版
  Uses a custom regex parser to render inline formulas ($e^{i\pi}$) and block equations ($$e^{i\pi} + 1 = 0$$) into high-performance, scrollable cyberpunk HTML math blocks.
• Multimodal Visual Attachments / 多模態附件
  Supports dragging/uploading images (.webp, .jpg, .png) to visual LLMs, or text documents (.py, .js, .txt) which are parsed and injected directly into conversation prompts.
• Speech Integration / 雙語語音串接
  Utilizes the Web Speech API to support microphone speech-to-text dictation and speech-to-text voice replies mapped dynamically to localized speakers.

📄 License / 授權條款

AetherAI Studio is open-source software licensed under the MIT License. 本專案採用 MIT 授權條款開放原始碼。