ai-code-switcher

Switch between profiles for AI coding tools and launch the right CLI with the right environment.

ai-code-switcher gives each profile its own provider, endpoint, credentials, proxy, and default arguments. The code-ai command then starts Claude Code, OpenAI Codex, or Grok using that profile.

Features

• Manage multiple profiles from one command-line interface
• Support API-key and isolated login/credential profiles
• Configure a custom endpoint and proxy for each profile
• Forward arguments to the underlying AI CLI
• Set per-profile default arguments such as --model
• Upgrade the supported AI CLIs through npm
• Optional Windows and macOS desktop launcher

Installation

Install the CLI from PyPI:

python -m pip install ai-code-switcher

For local development, install the checkout in editable mode:

python -m pip install -e .

The optional desktop launcher is installed with:

python -m pip install "ai-code-switcher[desktop]"

Requirements:

• Python 3.9 or later
• At least one supported AI CLI installed and available on PATH
• Node.js and npm for code-ai upgrade
• Windows or macOS for the optional desktop launcher

Quick start

Create a profile interactively, then launch it:

code-ai add
code-ai list
code-ai show my-claude
code-ai run my-claude

Arguments after the profile name are forwarded to the underlying CLI:

code-ai run my-claude -p "Review this repository"
code-ai run my-codex --model o3
code-ai run my-grok --help

Useful commands:

code-ai list                         List configured profiles
code-ai add                          Add a profile interactively
code-ai show <profile>               Show profile details
code-ai remove <profile>             Remove a profile
code-ai upgrade                      Upgrade Claude, Codex, and Grok CLIs
code-ai --version                    Show the installed version
code-ai --help                       Show command help

The supported AI CLIs are:

• Claude Code: @anthropic-ai/claude-code
• OpenAI Codex: @openai/codex
• Grok: @xai-official/grok

Profiles and configuration

Profiles are stored in ~/.code-ai/config.yaml. Use API mode when the profile should provide a token or API key. Use login mode when the CLI should use an isolated credentials directory.

Example configuration:

profiles:
  claude-api:
    name: claude-api
    type: claude
    mode: api
    base_url: https://api.anthropic.com
    token: sk-ant-...
    default_args:
      - --model
      - claude-sonnet-4-5

  codex-login:
    name: codex-login
    type: codex
    mode: login
    credentials_path: ~/.codex-profiles/work
    proxy: http://127.0.0.1:7890

  grok-api:
    name: grok-api
    type: grok
    mode: api
    base_url: https://api.x.ai
    api_key: xai-...

Login profiles isolate credentials through the provider-specific environment variables used by the CLIs. API profiles set the corresponding endpoint and authentication variables for the selected provider.

Keep tokens and personal configuration out of source control.

Default launch arguments

default_args can be a YAML list or a single string:

profiles:
  claude-api:
    name: claude-api
    type: claude
    default_args: "--model claude-sonnet-4-5"

Profile defaults are appended after command-line arguments. This lets a profile pin a value such as --model when the CLI uses the last occurrence of a flag. Skip profile defaults for one launch with:

code-ai run --no-default-args claude-api --model claude-opus-4-5

Claude Buddy

Claude profiles can have a deterministic companion card generated from the profile name:

code-ai roll-claude-buddy claude-api
code-ai roll-claude-buddy claude-api --name capybara
code-ai roll-claude-buddy claude-api --type legendary

Desktop launcher

The optional desktop launcher provides a graphical interface for detected Claude, ChatGPT, and Codex desktop apps. It is supported on Windows and macOS.

python -m pip install "ai-code-switcher[desktop]"
code-ai desktop install       # create or recreate the shortcut
code-ai desktop run           # open the launcher
code-ai desktop uninstall     # remove the shortcut

Desktop settings are stored separately in ~/.code-ai/desktop.yaml. The launcher can check the system proxy and apply common or per-application environment variables; per-application values take precedence.

Development

Run the test suite and build the distribution locally with:

python -m pytest tests/ -q
python -m build

License

MIT