Aether — AI Prompt Optimizer

Transform rough ideas into perfect AI prompts. Powered by local Ollama models — your data stays private.

Features

• Vibe-to-Prompt: Type a rough idea like "create a login page" and get a fully structured, production-ready AI prompt
• 100% Local & Private: Runs entirely on your machine via Ollama — no data leaves your computer
• Multi-IDE Support: Works with Cursor, Windsurf, Claude Code, GitHub Copilot, and VS Code
• Smart Context Scanning: Automatically analyzes your project structure for context-aware prompts
• Agent Selector: Choose your target AI model (Claude, GPT, Gemini, Grok) for optimized prompt formatting
• Quality Scoring: Every generated prompt gets a quality grade (A+, A, B, C, D)
• Security Auditing: Prompts are checked for injection risks and sensitive data leaks
• One-Click Model Management: Download, switch, and manage Ollama models from the sidebar
• Send to Agent: Instantly send generated prompts to your IDE's AI assistant

Quick Start

Prerequisites

1. Ollama — Install and run locally
2. Python 3.10+ — For the Brain backend server

Setup

1. Install the Aether extension from the VS Code Marketplace
2. Open a project folder in your IDE
3. Run the command "Aether: Start Brain" from the Command Palette (Ctrl+Shift+P)
4. The extension will:
   • Start Ollama (if not running)
   • Install Python dependencies
   • Launch the Brain server on port 8420
5. Select a model in the Welcome screen (recommended: gemma3:4b)
6. Start typing your ideas!

How It Works

Your Idea → Aether Brain → Ollama (Local LLM) → Optimized Prompt → Your AI Agent

1. You type a rough idea in the Aether sidebar
2. Context Scanner analyzes your project (files, structure, tech stack)
3. Prompt Optimizer crafts a detailed, structured prompt using the local LLM
4. Quality Auditor scores the output and checks for security issues
5. One click sends the prompt to your AI agent (Cursor, Copilot, etc.)

Configuration

Recommended Models

Privacy

Aether is designed with privacy as a core principle:

• Zero cloud dependency — All processing happens locally
• No telemetry — We don't collect any usage data
• No API keys needed — Works entirely with local Ollama models
• Your code stays yours — Context scanning is local-only

Architecture

aether/
├── brain/              # Python backend (FastAPI)
│   ├── sslm_engine.py          # Main server & Ollama integration
│   ├── prompt_optimizer.py     # Prompt generation engine
│   ├── prompt_knowledge_base.py # Best practices & patterns
│   ├── context_scanner.py      # Project analysis
│   ├── security_auditor.py     # Security checks
│   └── config.py               # Configuration
└── extension/          # VS Code extension (TypeScript)
    └── src/
        ├── extension.ts            # Entry point
        ├── providers/SidebarProvider.ts  # UI
        ├── services/BrainClient.ts      # HTTP client
        └── utils/config.ts              # Settings

Commands

Troubleshooting

Brain server won't start

• Make sure Python 3.10+ is installed: python --version
• Make sure Ollama is installed: ollama --version
• Check if port 8420 is available

No models showing up

• Ensure Ollama is running: ollama serve
• Try pulling a model manually: ollama pull gemma2:2b

Extension not connecting

• Check the Brain server URL in settings
• The extension auto-discovers ports 8420–8429

License

MIT © AhmetKayraKama