claude-router

Pick an OpenRouter model, launch Claude Code with it. Also includes provider scripts for MiniMax, Z.ai, and local models via Ollama.

Install

# Clone and symlink all scripts
git clone https://github.com/burka/claude-router.git
ln -s "$(pwd)/claude-router/claude-router"  ~/.local/bin/claude-router
ln -s "$(pwd)/claude-router/claude-minimax"         ~/.local/bin/claude-minimax
ln -s "$(pwd)/claude-router/claude-minimax-highspeed" ~/.local/bin/claude-minimax-highspeed
ln -s "$(pwd)/claude-router/claude-zai"             ~/.local/bin/claude-zai
ln -s "$(pwd)/claude-router/claude-ollama" ~/.local/bin/claude-ollama

# Set API keys in ~/.env (not needed for Ollama)
echo "OPENROUTER_API_KEY=sk-or-..."  >> ~/.env
echo "MINIMAX_API_KEY=..."           >> ~/.env
echo "ZAI_API_KEY=..."               >> ~/.env

Requires curl, jq, and optionally fzf for interactive picking.

Usage

claude-router                        # Interactive picker (fzf)
claude-router auto                   # Best free model
claude-router --free                 # Pick from free models only
claude-router --list                 # Browse available models
claude-router openai/gpt-4o          # Launch with specific model
claude-router auto -- -p "hello"     # Pass args to claude

claude-minimax                       # Launch with MiniMax MoE
claude-zai                           # Launch with Zhipu GLM (Z.ai token plan)
claude-ollama                        # Local Ollama shorthand (default: gemma4:26b)
claude-ollama qwen3.5:35b-a3b       # Local Ollama with specific model

# Or use Ollama's built-in Claude Code integration directly:
ollama launch claude --model gemma4:26b
ollama launch claude --model qwen3.5:35b-a3b

MiniMax Model Selection

By default claude-minimax uses MiniMax-M2.7. You can override with the MINIMAX_MODEL env var:

# ~/.env
MINIMAX_MODEL=MiniMax-M2.7-highspeed   # Optimized high-speed variant

Or via environment:

MINIMAX_MODEL=MiniMax-M2.7-highspeed claude-minimax

claude-minimax-highspeed is a thin wrapper that defaults to MiniMax-M2.1-Fusion — use it when you want the optimized variant without setting the env var.

Default Parameters

Set default claude CLI args via environment variable or ~/.env:

# ~/.env
CLAUDE_DEFAULT_PARAMS="--permission-mode auto"
CLAUDE_DEFAULT_PARAMS="--team"
CLAUDE_DEFAULT_PARAMS="-t"

Or via environment:

CLAUDE_DEFAULT_PARAMS="--permission-mode auto" claude-router openai/gpt-4o

User-specified args always override CLAUDE_DEFAULT_PARAMS.

Default Wrapper

Wrap the claude launch command with nice, taskset, prlimit, etc. Set via ~/.env or environment:

# ~/.env — pin to cores 2-7, set memory limits
CLAUDE_DEFAULT_WRAPPER="nice -n 20 taskset -c 2-7 prlimit --as=128000000000 --rss=64000000000"

# Via environment
CLAUDE_DEFAULT_WRAPPER="echo running:" claude-router -p "hello"

Local Models with Ollama

Run Claude Code against local models — no API key needed, fully offline.

Native way (Ollama ≥ 0.9)

ollama launch claude --model gemma4:26b
ollama launch claude --model qwen3.5:35b-a3b

This is Ollama's built-in Claude Code integration — it handles download, serving, and env setup for you.

Shorthand via claude-ollama

claude-ollama is a thin wrapper (~25 lines) that does the same thing with auto-pull:

claude-ollama                        # Default: gemma4:26b
claude-ollama qwen3.5:35b-a3b       # Use Qwen 3.5
claude-ollama gemma4:26b -p "hi"    # Pass args to claude

OLLAMA_HOST=http://192.168.1.100:11434 claude-ollama  # Remote host

How it works

1. claude-router fetches models from the OpenRouter API (cached daily in ~/.cache/claude-router/)
2. Filters for text models with tool support
3. Sets ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN, and model env vars, then launches claude
4. claude-minimax and claude-zai set their provider's endpoint and auth directly
5. claude-ollama connects to a local Ollama instance via its OpenAI-compatible API (/v1)

License

MIT