claunch

Claude Code smart launcher with fzf model switcher.

Run a different AI model in every terminal window — simultaneously, without conflicts.
One window on Claude Opus, another on MiniMax, another on DeepSeek. Each session is fully isolated.

中文文档

---

Why claunch

Most setups force you to pick one model globally. claunch lets you open multiple terminal windows, each running a different provider or model at the same time — no config files to swap, no environment leaking between sessions.

How it works: claunch injects model credentials as process-level environment variables (env KEY=VAL claude ...). Each terminal process has its own environment, so switching models in one window never affects another. Switch models per-window, per-task, per-context.

Features

• Per-window model isolation — each terminal session runs its own model, completely independent
• ca --new — pick any model via fzf before launching
• ca — launch with the last-used model in this window
• ca --list — browse models interactively: Enter launch, e edit, Del delete
• Model management: add, remove, edit models without touching JSON files
• Background version check — notifies when an upgrade is available
• Bilingual UI: English and Chinese (ca --lang zh)
• All claude flags pass through (e.g. ca --continue, ca --resume <id>)
• Restores terminal state (p10k, Starship, Pure, and other prompt frameworks) cleanly after exit

Requirements

• Claude Code (claude CLI)
• Homebrew (for auto-installing jq and fzf)
• zsh

Install

bash <(curl -fsSL https://raw.githubusercontent.com/k186/claunch/main/install.sh)
source ~/.zshrc

jq and fzf are installed automatically via Homebrew if missing.

Or clone and install locally:

git clone https://github.com/k186/claunch ~/claunch
zsh ~/claunch/install.sh
source ~/.zshrc

Usage

ca                      # launch with current model
ca --new                # pick model with fzf, then launch
ca --continue           # resume last session
ca --new --resume <id>  # pick model + resume session

All claude flags pass through verbatim after ca.

Model management

ca --list               # browse models (Enter=launch, e=edit, Del=delete)
ca --add                # add a new model (interactive wizard)
ca --remove             # remove a model (fzf picker)
ca --current            # show which model is active in this window

ca --list opens an interactive fzf panel with a live preview of each model's configuration. Press Enter to launch, e to edit, Del to delete (with confirmation).

Other commands

ca --update            # upgrade claunch (models.json is never modified)
ca --lang [en|zh]       # show or set the UI language
ca --help               # show all commands

claunch checks for updates in the background on every launch and prints a notice if a newer version is available.

Configuration

~/.claude/models.json is created from models.example.json on first install. You can also manage models interactively with ca --add, ca --remove, and ca --list.

{
  "name": "claunch",
  "lang": "en",
  "models": [
    {
      "name": "Claude Opus 4.7",
      "model": "claude-opus-4-7",
      "env": {}
    },
    {
      "name": "MiniMax-M2.7",
      "model": "",
      "env": {
        "ANTHROPIC_BASE_URL": "https://api.minimaxi.com/anthropic",
        "ANTHROPIC_AUTH_TOKEN": "your-api-key",
        "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1",
        "ANTHROPIC_MODEL": "MiniMax-M2.7"
      }
    },
    {
      "name": "DeepSeek V4 Pro (1M)",
      "model": "",
      "env": {
        "ANTHROPIC_BASE_URL": "https://api.deepseek.com/anthropic",
        "ANTHROPIC_AUTH_TOKEN": "your-api-key",
        "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1",
        "CLAUDE_MAX_CONTEXT_WINDOW": "1000000",
        "ANTHROPIC_MODEL": "deepseek-v4-pro[1m]"
      }
    }
  ]
}

Field reference:

Field	Description
name	Display name shown in fzf
model	Passed as --model to claude. Leave "" to use the provider's default via env vars
env	Environment variables injected per-session (API keys, base URLs, etc.)

For third-party providers (MiniMax, DeepSeek, etc.), set:

• ANTHROPIC_BASE_URL — provider's Anthropic-compatible API endpoint
• ANTHROPIC_AUTH_TOKEN — your API key
• ANTHROPIC_MODEL — model name as the provider expects it
• CLAUDE_MAX_CONTEXT_WINDOW — optional, e.g. "1000000" for 1M context
• CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC — set to "1" for third-party providers

License

MIT