llm-gateway

LLM Gateway

Use GitHub Copilot, Kimi, MiniMax, GLM, Qwen models as Claude Code's agent model.

Install

curl -fsSL https://raw.githubusercontent.com/SEUFSG/llm-gateway/master/install.sh | bash

Requirements: bun (or node), git, claude CLI.

Setup

# Interactive — authenticate all providers at once
llm-auth setup

# Or one by one
llm-auth login copilot            # GitHub OAuth device flow
llm-auth login kimi --key sk-xxx
llm-auth login qwen --key sk-xxx
llm-auth login minimax --key xxx
llm-auth login glm --key xxx.xxx

# Check status
llm-auth status

# List all available models
llm-auth models

Usage

# Start Claude Code with a specific model
claude --model copilot/claude-opus-4.7
claude --model copilot/gpt-5.4
claude --model copilot/gpt-4o
claude --model kimi/kimi-latest
claude --model qwen/qwen-max
claude --model glm/glm-4-plus
claude --model minimax/MiniMax-Text-01

Model IDs always use provider/model format to avoid conflicts between providers.

Supported Providers

Provider	Auth Method	Example Models
GitHub Copilot	OAuth Device Flow	claude-opus-4.7, claude-sonnet-4.6, gpt-5.4, gpt-4o, gemini-2.5-pro, grok-code-fast-1
Moonshot Kimi	API Key	kimi-latest, kimi-thinking-preview, moonshot-v1-128k
Alibaba Qwen	API Key	qwen3-235b-a22b, qwen-max, qwen-plus, qwen2.5-coder-32b-instruct
MiniMax	API Key	MiniMax-Text-01, abab6.5-chat
Zhipu GLM	API Key	glm-z1-preview, glm-4-plus, glm-4-flash

How It Works

1. llm-auth manages credentials in ~/.llm-gateway/credentials.json
2. Proxy server (localhost:3456) translates Anthropic API format to each provider's API
3. ANTHROPIC_BASE_URL points Claude Code to the proxy
4. SessionStart hook auto-starts the proxy when Claude Code launches

Uninstall

claude mcp remove llm-gateway -s user
rm -rf ~/.local/share/llm-gateway ~/.local/bin/llm-auth ~/.llm-gateway
# Remove ANTHROPIC_BASE_URL and SessionStart hook from ~/.claude/settings.json