🧠 Cortex

🚀 What is Cortex?

Cortex solves the "amnesia" problem for AI coding assistants. Your AI forgets everything when you close the window. Cortex doesn't.

It's a universal, persistent memory layer that lets your AI (Copilot, Claude, Cursor, etc.) remember:

• 🏛️ Architecture Decisions: "We use Feature-Sliced Design"
• 💡 Code Patterns: "All React components must use functional style"
• 🔧 Configuration: "Production uses AWS RDS with specific flags"
• 🚫 Exclusions: "Never touch the legacy payment module"

Unlike cloud-based solutions, Cortex stores everything locally on your machine (~/.cortex/memories.db). Your code context never leaves your computer.

💡 Stop teaching your AI the same things over and over.

✨ Key Features

1. 🧠 AI-Powered Project Scanner (Two-Pass)

Cortex doesn't just "read files". It performs a semantic analysis of your entire project:

1. Pass 1 (Tree Scan): The AI analyzes your file structure to understand the project's topology.
2. Pass 2 (Extraction): It selectively reads key files to extract "Facts", "Decisions", and "Patterns", storing them as structured memories.

2. 🌐 Universal MCP Support

Built on the Model Context Protocol (MCP), Cortex works with any modern AI tool:

• VS Code / Cursor / Windsurf: Native integration.
• Claude Desktop: Full memory access.
• JetBrains / Neovim: Via MCP stdio.
• Command Line Agents: Works with Goose, Zed, and others.
• See full list of supported tools

3. 🔐 Privacy & Local-First

• 100% Local Storage: Memories are stored in SQLite (~/.cortex/memories.db) on your machine.
• Privacy Guard: (Coming Soon) Automatic PII redaction before context injection.
• Project Isolation: Memories are automatically scoped to the specific git repository you are working on.

4. 🔎 Hybrid Search

Cortex uses a hybrid approach for finding context:

• FTS5 (Full Text Search): Fast, exact matching for keywords/symbols.
• Vectors (Semantic Search): (Optional) If you provide an OPENAI_API_KEY, Cortex enables semantic understanding to find related concepts even if keywords don't match.

⚡ One-Command Install (Recommended)

Cortex provides a universal installer that automatically configures Cursor, Windsurf, Claude Code, Gemini Code Assist, VS Code, and Zed in one go.

npx @ecuabyte/cortex-cli setup

This will:

1. Detect your installed AI editors.
2. Configure them to use the Cortex MCP Server.
3. Scan your current project for initial context.

Manual Installation

If you prefer to configure manually or use a specific editor:

1. Install CLI Globally

npm install -g @ecuabyte/cortex-cli
# or
bun add -g @ecuabyte/cortex-cli

2. Configure Specific Editor

# For Cursor
cortex install --editor cursor

# For Claude Desktop
cortex install --editor claude-desktop

3. Manual Config Generation

If you need the raw JSON configuration:

npx -y @ecuabyte/cortex-mcp-server generate-config

goose configure mcp add cortex "npx -y @ecuabyte/cortex-mcp-server"

📦 Architecture

Cortex allows you to compose 5 core primitives:

// 1. STORE: Save items (facts, decisions, code)
await cortex.store({ content: "Use Zod for validation", type: "decision" });

// 2. ROUTE: Find context relevant to a generic query
const context = await cortex.route({ task: "Implement user login" });

// 3. GET: Retrieve specific memories
const memories = await cortex.get({ type: "code", tags: ["auth"] });

// 4. GUARD: (Beta) Sanitize output
const safe = await cortex.guard(content);

// 5. SCAN: Analyze project structure
await cortex.scan();

🗺️ Roadmap & Status

🤝 Contributing

We welcome contributions! Please see CONTRIBUTING.md.

git clone https://github.com/EcuaByte-lat/Cortex.git
cd Cortex
bun install
bun run build

📄 License

MIT License © EcuaByte