Kama - AI Prompt Optimizer

Transform rough ideas into perfect AI prompts. 100% local, no API keys needed — your data stays private.

What is Kama?

Kama turns your rough ideas into production-ready AI prompts. Type something like "create a login page with OAuth" and get a structured, context-aware prompt optimized for your target AI — all running 100% locally on your machine.

No API keys. No cloud. No data leaves your computer.

Features

Core

• Vibe-to-Prompt — Type a rough idea, get a fully structured AI prompt in seconds
• 100% Local & Private — Runs entirely on your machine using a local GGUF model (~1.3 GB)
• Plug & Play — One-click install. Brain server + model download automatically on first use
• Streaming Output — Watch prompts being generated token-by-token in real time
• Smart Context — Automatically scans your project structure, languages, and tech stack

Prompt Tools

• Prompt History & Favorites — Full history with search, star favorites, delete items. Persistent across sessions (up to 100 entries)
• Prompt Chains — Chain prompts together for multi-step workflows. Click the Chain button on any prompt, then your next vibe builds on the previous one
• Active File Context — Kama reads your currently open file and feeds it to the AI for more relevant prompts
• Keyboard Shortcut — Ctrl+Shift+K generates a prompt from your current file or selection
• Onboarding Tutorial — 4-slide walkthrough for first-time users

Agent & Model Support

• Multi-IDE Support — Works with Cursor, Windsurf, Claude Code, GitHub Copilot, and VS Code
• Agent Selector — Pick your target AI (Claude, GPT, Gemini, Grok, Codex, o3) for optimized prompt formatting
• Model Management — Download, switch, and manage local GGUF models from the sidebar
• Send to Agent — One click sends the generated prompt directly to your IDE's AI assistant

Quality & Security

• Quality Scoring — Every prompt gets a quality grade (A+ to D)
• Security Auditing — Prompts checked for injection risks and sensitive data leaks
• Prompt Sanitization — Dangerous patterns automatically removed
• Rate Limiting — Per-client, in-memory rate limiting on all endpoints
• Request Size Limits — Bodies over 1 MB are rejected to prevent abuse

Quick Start

Prerequisites

• Python 3.10+ — Download (check "Add to PATH" during install)
• 8 GB RAM minimum (16 GB recommended)

That's it. No Ollama, no Docker, no external services needed.

Setup

1. Install Kama from the VS Code Marketplace
2. Open any project folder
3. Click the Kama icon in the Activity Bar (left sidebar)
4. Kama will automatically:
   • Start the Brain server in the background
   • Create a Python virtual environment
   • Install Python dependencies
   • Download a local AI model (~1.3 GB, one-time)
5. Start typing your ideas!

How It Works

Your Idea -> Kama Brain -> Local LLM (GGUF) -> Optimized Prompt -> Your AI Agent

1. You type a rough idea in the Kama sidebar
2. Security Auditor checks input for injection/credential leaks
3. Context Scanner analyzes your project (files, structure, tech stack)
4. Active File context from your currently open editor is included
5. Local LLM generates a detailed, AI-optimized prompt (streamed token-by-token)
6. Quality Scorer grades the output (A+ to D)
7. Prompt Sanitizer strips dangerous patterns
8. One click sends the prompt to your AI agent (Cursor, Copilot, Claude, etc.)

Configuration

Available Models

Models download automatically to ~/.kama/models/ on first use. Switch models from the sidebar.

Commands

Privacy & Security

• Zero cloud dependency — All processing happens locally
• No telemetry — We don't collect any usage data
• No API keys — Works entirely with a bundled local model
• Your code stays yours — Context scanning never leaves your machine
• Models stored locally — In ~/.kama/models/, you own everything
• CORS restricted — Brain API only accepts requests from localhost
• Input sanitization — Injection and credential leak detection on every request
• Download verification — Model files are SHA-256 verified after download

Architecture

kama/
├── brain/                        # Python backend (bundled with VSIX)
│   ├── sslm_engine.py            # FastAPI server + endpoints
│   ├── llm_backend.py            # Local GGUF model management
│   ├── prompt_optimizer.py       # AI-specific prompt builder
│   ├── prompt_knowledge_base.py  # Community prompt patterns
│   ├── context_scanner.py        # Project analysis
│   ├── security_auditor.py       # Security gate
│   ├── hardware_profiler.py      # Hardware-aware model recommendations
│   └── config.py                 # Environment-driven configuration
└── extension/                    # VS Code extension (TypeScript)
    └── src/
        ├── extension.ts               # Entry point + auto-start + IDE detection
        ├── providers/SidebarProvider.ts # Webview UI + streaming + history
        ├── services/BrainClient.ts     # HTTP/SSE client
        └── utils/config.ts            # Settings

Troubleshooting

Brain server won't start

• Make sure Python 3.10+ is installed: python --version
• Check if port 8420 is available
• Try restarting VS Code
• Check the "Kama Brain" terminal in VS Code for errors

Model download stuck

• Check your internet connection (models download from HuggingFace)
• Models are cached in ~/.kama/models/ — delete the folder to re-download

Extension shows "Offline"

• The Brain auto-starts on activation. Wait 10-15 seconds for first-time setup
• Click "Start Brain Server" in the offline banner

Contributing

1. Fork the repo
2. Clone your fork and open the extension/ folder in VS Code
3. npm install in the extension/ directory
4. Press F5 to launch Extension Development Host
5. Make changes and submit a PR

Support

If you find Kama useful, consider supporting the project:

License

MIT — Made by Ahmet Kayra Kama