Claudex

Claudex is a terminal AI coding assistant that runs directly in your terminal — no gateways, no proxies, no external servers. Just you, your terminal, and your choice of AI provider.

Quick Start

npm install -g claudex-ai
claudex

Then press /model to pick your provider and authenticate.

Other install options

# Linux/Mac (curl)
curl -fsSL https://raw.githubusercontent.com/JaredBautist/CLAUDEX/main/scripts/install.sh | bash
claudex

# GitHub direct
npm install -g github:JaredBautist/CLAUDEX
claudex

Features

Multi-Provider — Claude, OpenAI, and Gemini. Switch anytime with /model.

claudex use claude claude-sonnet-4-6
claudex use openai gpt-4o
claudex use gemini gemini-2.5-pro

30+ Built-in Skills — Invoke with /<skill-name>. The model auto-selects them based on your task.

| Category | Skills | |----------|--------| | Frontend | codex-frontend-master, frontend-design, v0-frontend, ui-ux-pro-max | | Code Quality | verify, review, test, refactor, simplify, self-test | | Architecture & Docs | spec, architecture, docs, commit | | Backend & Infra | api-design, database, deploy, backend-security | | Productivity | debug, onboard, instructions, session, batch, stuck, remember, learn, token-saver | | AI & Providers | ai-provider, skillify, keybindings, update-config |

Spec-Driven Development — Define requirements, design, and tasks in .claudex/SPEC.md. The model reads and updates them as you work.

Web Search & Fetch — Built-in web search and URL fetching. The model uses them automatically when it needs current information.

Session Persistence — Save and restore session state across terminal sessions with /session save and /session resume.

Custom Instructions — Add project-specific rules in .claudex/instructions/. They load every turn automatically.

Adaptive Learning — Claudex learns your preferences over time. Use /learn to manage memory, skill preferences, and suggestions.

Token Optimization — /token-saver with 4 modes (auto, eco, turbo, budget) to control token consumption. Per-message usage display shows in:X out:Y for every response.

Usage

CLI Commands

claudex                    # Open the TUI
claudex chat "prompt"      # One-shot query, no TUI
claudex auth <provider>    # Configure API key
claudex use <provider> <model>  # Switch model
claudex -p openai -m gpt-4o-mini chat "hi"  # Override

Inside the TUI

/model                    # Choose provider & model
/effort normal            # Set response depth

/spec init                # Start project spec
/review                   # Code review
/test                     # Write & run tests
/refactor                 # Safe refactoring
/commit                   # Conventional commit
/self-test                # Run Claudex's own checks
/session save             # Save state
/learn                    # Manage learning & preferences
/token-saver eco          # Optimize token usage
/v0-frontend              # Vercel-v0-style frontend

Requirements

• Node.js 20+
• An API key or OAuth for your chosen provider

Configuration

Config is stored outside the repo:

~/.config/claudex/direct-providers.json
~/.config/claudex/claude-runtime/

Or use environment variables:

export ANTHROPIC_API_KEY="sk-..."
export OPENAI_API_KEY="sk-..."
export GEMINI_API_KEY="..."

Provider Support

| Provider | Models | Auth | |----------|--------|------| | Anthropic Claude | Claude 4, Claude Sonnet 4, Claude Haiku 4 | API key / OAuth | | OpenAI | GPT-4o, GPT-4o-mini, GPT-5.x | API key / OAuth | | Google Gemini | Gemini 2.5 Pro, Gemini 2.5 Flash | API key |

Development

git clone https://github.com/JaredBautist/CLAUDEX.git
cd CLAUDEX
npm install
npm start

How It Differs

Claudex is a direct-provider TUI. Unlike tools that require a proxy server or gateway:

• Your API calls go directly to the provider — nothing in between
• You own your keys and your data
• No external dependencies beyond Node.js and npm

The skills system is inspired by Claude Code but extended with Spec-Driven Development, session persistence, structured project memory via .claudex/, adaptive learning, and per-message token tracking.

Project Structure

.claude/skills/             # Official Claude Code skills (auto-discovered)
.claudex/                   # Project specs, instructions, session state
  SPEC.md                   #   Requirements, design, tasks
  instructions/             #   Custom project rules
  session.json              #   Current session state
  learning/                 #   Adaptive learning data
    state.json              #     Skill usage, preferences, memories
src/skills/bundled/         # Built-in TypeScript skills (30+)
skillpacks/                 # Curated skill packs

License

MIT

Acknowledgments

Built on ideas from Claude Code, Claude Dev, and the open-source AI tooling ecosystem.