🌟 Why Catui?

"The only AI coding assistant that truly learns from you"

Catui isn't just another AI coding tool. It's a terminal-native AI agent with persistent memory and evolving personality — designed for developers who live in the terminal.

What Makes It Different?

| | Other Tools | Catui | |---|---|---| | Memory | ❌ Starts fresh every session | ✅ Remembers your projects, preferences, and coding style | | Personality | ❌ Generic responses | ✅ Evolves a unique personality based on your interactions | | Terminal Native | ❌ GUI wrappers or plugins | ✅ Pure TUI built for terminal workflows | | Model Freedom | ❌ Vendor lock-in | ✅ 10+ providers, switch instantly | | Offline Ready | ❌ Cloud dependent | ✅ Local models via Ollama |

✨ Features

🧠 Persistent Memory (NanoMem)

Your projects have context. So should your AI.

• Project Knowledge — API endpoints, database schemas, architecture decisions
• Error Patterns — Remembers bugs and their solutions
• User Preferences — Coding style, naming conventions, framework choices
• Smart Retrieval — Automatically recalls relevant context when needed

💫 AI Personality Evolution (NanoSoul)

Your AI assistant develops its own character.

• Big Five Traits — Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism
• Coding Style — Verbosity, abstraction level, safety margins
• Domain Expertise — Frontend, Backend, DevOps, AI/ML specializations
• Emotional States — Confidence, curiosity, flow state

After 50 sessions, your Catui will code differently than anyone else's.

🎨 Beautiful TUI

A terminal interface that feels alive.

• Three Themes — Dark, Light, and Warm (eye-friendly)
• Smooth Animations — Breathing catui loader
• Intuitive Navigation — Vim-like keybindings, fuzzy search
• Real-time Streaming — Watch the AI think in real-time

🔌 MCP Protocol Support

Extend with the Model Context Protocol.

Built-in tools include:

• 📁 Filesystem operations
• 🌐 HTTP requests (Fetch)
• 🎭 Browser automation (Puppeteer)
• 🗄️ Database queries (SQLite)
• 🔧 Git operations

🌐 Multi-Model Support

Use the best model for each task.

• 🇨🇳 Alibaba DashScope / Token Plan — Qwen, GLM, MiniMax, DeepSeek text models
• 🤖 OpenAI — GPT-4, GPT-3.5
• 💬 Anthropic — Claude 3 Opus/Sonnet/Haiku
• 🔍 Google — Gemini Pro/Ultra
• 🏠 Local — Ollama (Qwen, Llama, Mistral, etc.)

🚀 Quick Start

Installation

npm install -g catui-agent

First Run

catui

1. Select your model — Choose from available providers
2. Enter API key — Securely stored in ~/.catui/agents/default/auth.json
3. Start coding — Just type what you want to build

Example Session

You: Create a JWT authentication middleware for Express

AI: [Analyzing project structure...]
    [Creating auth.middleware.ts...]
    [Adding TypeScript types...]
    [Writing tests...]
    
    ✅ Done! Created:
    - src/middleware/auth.middleware.ts
    - src/types/auth.d.ts
    - tests/auth.middleware.test.ts
    
    Features:
    • JWT verification with RS256
    • Refresh token rotation
    • Role-based access control
    • Rate limiting integration

🎮 Usage

Interactive Mode

catui                    # Start new session
catui -c                 # Continue last session
catui -r                 # Resume from history
catui -m qwen-max        # Use specific model

Print Mode (Scripts)

# Single query
catui -p "Refactor this to use async/await"

# Pipe input
cat bug-report.md | catui -p "Analyze this bug"

# Chain commands
catui -p "Generate API" | catui -p "Write tests for it"

Slash Commands

| Command | Description | |---------|-------------| | /model | Switch AI model | | /thinking | Adjust reasoning depth | | /fork | Branch conversation | | /tree | Browse session history | | /memory | View project memories | | /soul | Check AI personality | | /settings | Configure preferences | | /export | Save as HTML |

📊 Comparison

| Feature | Catui | Cursor | Claude Code | Aider | |---------|:----------:|:------:|:-----------:|:-----:| | Terminal Native | ✅ | ❌ | ✅ | ✅ | | Persistent Memory | ✅ | ❌ | ❌ | ❌ | | AI Personality | ✅ | ❌ | ❌ | ❌ | | Session Branching | ✅ | ✅ | ✅ | ❌ | | Multi-Model | ✅ | ✅ | ❌ | ✅ | | MCP Support | ✅ | ❌ | ✅ | ❌ | | Offline Mode | ✅ | ❌ | ❌ | ✅ | | Chinese Optimized | ✅ | ❌ | ❌ | ❌ |

🏗️ Architecture Philosophy

Catui is built on three pillars:

┌─────────────────────────────────────────┐
│           🧠 COGNITIVE LAYER            │
│    (Memory + Personality + Context)     │
├─────────────────────────────────────────┤
│           🔧 TOOL LAYER                 │
│  (File Ops + Bash + Search + MCP)       │
├─────────────────────────────────────────┤
│           🎨 INTERFACE LAYER            │
│       (TUI + Themes + Keybindings)      │
└─────────────────────────────────────────┘

Design Principles:

• Terminal First — No Electron, no browser, pure terminal
• Privacy First — Local storage by default; opt-in diagnostics/eval telemetry only when explicitly configured
• Extensible — Plugin system for tools, themes, and behaviors
• Fast — Sub-second startup, instant response

📚 Documentation

• Installation Guide
• Configuration
• Memory System
• MCP Guide
• Keybindings
• Extensions
• Run Trace, Replay, and Harness Eval

🌍 Community

• 💬 Discussions
• 🐛 Issues
• 📝 Changelog

📄 License

MIT © Catui Agent

⬆ Back to Top

Built with ❤️ for terminal dwellers everywhere