Claudex

Use Claude Code with any AI provider — Grok, ChatGPT, DeepSeek, Gemini, Kimi, GLM, Ollama, Azure, Groq, Mistral, and 20+ more.

Get Started View on GitHub

Features

Translation Proxy

Automatic Anthropic-to-OpenAI protocol translation. Use Claude Code with 20+ providers including Grok, ChatGPT, DeepSeek, Gemini, Kimi, GLM, Azure OpenAI, Groq, Mistral, Together AI, Perplexity, Cerebras, and any OpenAI-compatible endpoint.

Circuit Breaker

Per-profile circuit breaker with automatic failover to backup providers. Configurable failure thresholds, recovery timeouts, and OAuth 401 retry with token refresh.

Smart Routing

Intent-based auto-routing via local classifier. Automatically maps code, analysis, creative, search, and math tasks to the optimal provider profile.

Context Engine

Conversation compression, cross-profile context sharing, and local RAG with embeddings. Enhance your AI interactions with richer context.

TUI Dashboard

Real-time terminal dashboard showing profile health, request metrics, logs, and quick-launch capabilities. Full keyboard navigation.

Self-Update

One command to check and install the latest version directly from GitHub Releases. Stay up to date effortlessly.

OAuth Subscriptions

Use existing Claude, ChatGPT, Google, Qwen, Kimi, GitHub Copilot, or GitLab Duo subscriptions. Supports browser PKCE, device code flow, and credential chain from native CLIs.

Configuration Sets

Install and manage reusable bundles of CLAUDE.md rules, skills, and MCP servers via claudex sets add. Versioned manifests with conflict detection.

Terminal Hyperlinks

OSC 8 clickable hyperlinks in terminal output. Auto-detects terminal support (iTerm, WezTerm, Kitty, Ghostty, VS Code, Windows Terminal) with PTY proxy mode.

Model Slot Mapping

Map Claude Code’s /model switcher (haiku, sonnet, opus) to any provider model via [profiles.models]. Per-profile model configuration with environment variable overrides.

Non-Interactive Mode

One-shot execution with claudex run <profile> "prompt" --print. Ideal for CI/CD pipelines and scripted automation.

Tool Name Compatibility

Auto-truncates tool names longer than 64 characters for OpenAI compatibility, with transparent roundtrip restoration on responses.

Supported Providers

Category	Providers
Direct Anthropic	Anthropic, MiniMax
OpenAI Compatible	Grok (xAI), OpenAI, DeepSeek, Kimi/Moonshot, GLM (Zhipu), OpenRouter, Groq, Mistral, Together AI, Perplexity, Cerebras
Cloud Platforms	Azure OpenAI, Google Vertex AI
OAuth Subscriptions	Claude Max/Pro, ChatGPT Plus/Pro (Codex), Google Gemini, GitHub Copilot, GitLab Duo, Qwen, Kimi
Local Models	Ollama, vLLM, LM Studio, llama.cpp
Via LiteLLM	AWS Bedrock, and any LiteLLM-supported provider

Quick Install

curl -fsSL https://raw.githubusercontent.com/StringKe/claudex/main/install.sh | bash

How It Works

claudex run grok
    │
    ├── Start proxy → 127.0.0.1:13456
    │
    └── exec claude with:
        ANTHROPIC_BASE_URL=http://127.0.0.1:13456/proxy/grok
        ANTHROPIC_AUTH_TOKEN=<session-token>
        ANTHROPIC_MODEL=grok-3-beta

The proxy transparently translates between Anthropic and OpenAI protocols, so Claude Code works seamlessly with any supported provider. When using OAuth profiles, ANTHROPIC_AUTH_TOKEN (not ANTHROPIC_API_KEY) is set to avoid conflicts with Claude’s own subscription login.