Skip to main content
CC Switch is a desktop app that writes provider settings into the config files of Claude Code, Codex, Gemini CLI, and other coding tools, so you switch providers with one click instead of editing each file by hand.

​
Install CC Switch

CC Switch is open source under the MIT license and runs on macOS 12 or later, Windows, and Linux.
  • macOS: run brew install --cask cc-switch, or download the .dmg from the GitHub releases page.
  • Windows: download the MSI installer or the portable build.
  • Linux: download the .deb, .rpm, or AppImage package. It needs glibc 2.35+ and WebKitGTK 4.1.
On first launch, CC Switch imports your current Claude Code, Codex, and Gemini CLI settings as a provider named default, so nothing you set up before is lost.

​
Add LaoZhang API to all three tools

Claude Code needs a key in the Claude Code token group, while Codex and Gemini CLI need a key in the default group. A universal provider holds only one key, so add Claude Code as its own provider and let a universal provider cover Codex and Gemini CLI.
1

Create two tokens

In the LaoZhang API console, create two tokens with the Usage first billing mode and copy both keys:
  • One in the Claude Code group, for Claude Code.
  • One in the default group, for Codex and Gemini CLI.
2

Add a provider for Claude Code

Select Claude Code in CC Switch’s sidebar, click Add Provider (+) at the top right, and choose Custom Configuration. Enter a name such as “LaoZhang API”, the endpoint https://api.laozhang.ai, and your Claude Code-group key, then fill in Model Mapping with IDs from Choose models. The full field list is under Claude Code below. Click Add.
Fill the Haiku row with a model LaoZhang API serves, such as claude-haiku-5-5. Claude Code runs background tasks on the Haiku model, and claude-haiku-4-5-20251001 or an empty row makes those requests fail with a 503. If you don’t use Claude, enter deepseek-v4-flash or the same model as the other rows.
3

Open the universal provider presets

Select Codex or Gemini CLI in the sidebar and click Add Provider at the top right. In the category list on the left, under Across apps, open Universal providers and choose 自定义网关 (Custom gateway). This preset name appears in Chinese even when the interface is in English.
4

Enter the address and key

  • Name: anything you like, such as “LaoZhang API”
  • API URL: https://api.laozhang.ai
  • API Key: your default-group key
  • Enabled Apps: turn off Claude Code and leave Codex and Gemini CLI on
Enter the bare address once. CC Switch adds /v1 for Codex on its own, as shown in the table below this procedure.
5

Set the Codex and Gemini models

Under Model Configuration, the form has one group per tool. Set the Codex model (and reasoning effort, if you like) and the Gemini model to models LaoZhang API serves; the values in Choose models work as they are. Claude Code is off in this provider, so its fields aren’t used.
6

Add and switch

Click Add. CC Switch saves the provider and syncs it to Codex and Gemini CLI. Then, for each of the three tools, find the new LaoZhang API card in its provider list and click Switch. The card shows In Use when it’s active.Until you switch, each tool’s card for default is marked In Use. That provider holds the settings CC Switch imported from your machine on first launch.
  • Clicking Switch rewrites that tool’s config file.
  • Before the first rewrite, CC Switch backs up the original file to ~/.cc-switch/backups/live-first-write/.
  • Switching back to default restores your original setup.
7

Restart and test

Claude Code picks up the change right away. Codex and Gemini CLI read their config only at startup, so quit any running session and start a new one. Then send a test request from a terminal:
Each tool should print connected. Gemini CLI runs non-interactive prompts only in a trusted directory, so trust the directory in interactive mode first if it asks.
From one API URL, CC Switch writes these addresses: All three tools call LaoZhang API directly in their native protocols: Anthropic Messages, the OpenAI Responses API, and native Gemini. You don’t need CC Switch’s local routing.

​
Add a provider for one tool

Claude Code always uses this path. Use it for Codex or Gemini CLI too when you set up a single tool or need more fields than a universal provider shows: select the tool at the top of CC Switch, click Add Provider (+), and choose Custom Configuration. Fill in the fields below, click Add, then click Switch on the new card.

​
Claude Code

  • API Endpoint: https://api.laozhang.ai, with no trailing slash. CC Switch writes it to ANTHROPIC_BASE_URL.
  • API Key: your Claude Code-group key.
  • Auth Field: keep the default, ANTHROPIC_AUTH_TOKEN.
  • Upstream Format: keep Anthropic Messages (Native). The other formats need local routing, which LaoZhang API doesn’t require.
  • Model Mapping: fill in the requested model for the Sonnet, Opus, Fable, and Haiku rows, plus Subagent if you use it. Each tier row becomes an ANTHROPIC_DEFAULT_*_MODEL variable, such as ANTHROPIC_DEFAULT_FABLE_MODEL, and Subagent becomes CLAUDE_CODE_SUBAGENT_MODEL.
  • Default fallback model: the model Claude Code starts with, written to ANTHROPIC_MODEL.
Don’t leave the Haiku row empty. An empty row makes Claude Code send its original Claude model name, which LaoZhang API may not offer. Use claude-haiku-5-5, or deepseek-v4-flash if you don’t use Claude. To pick IDs from a list instead of typing them, click Fetch Models after entering the endpoint and key.

​
Codex

  • API Request URL: https://api.laozhang.ai/v1. CC Switch writes it to base_url under [model_providers.custom] in ~/.codex/config.toml.
  • API Key: your default-group key. CC Switch stores it in config.toml as experimental_bearer_token, not in auth.json.
  • Upstream Format: keep OpenAI Responses API (Native), which sets wire_api = "responses".
  • Default Model: gpt-6.1-sol, or another model from the table in Choose models.
Restart Codex after switching.

​
Gemini CLI

Enter https://api.laozhang.ai as the address, your default-group key, and a model such as gemini-3.8-flash. When you switch, CC Switch writes:
  • GOOGLE_GEMINI_BASE_URL, GEMINI_API_KEY, and GEMINI_MODEL to ~/.gemini/.env
  • the Gemini API key sign-in method and the model name to ~/.gemini/settings.json
Restart Gemini CLI after switching.

​
Check the endpoint

Click Manage & Test next to the request address to measure response times. Connectivity check only confirms that the address returns an HTTP response; it doesn’t check your key. Use the test commands above to confirm the key works.

​
Choose models

Enter the Claude Code IDs in Claude Code’s custom configuration, and the Codex and Gemini CLI IDs in the universal provider’s Model Configuration or in each tool’s custom configuration. Use the Claude column or the GLM-5.2 column for Claude Code, not a mix.
  • Fill the Fable row in Claude Code’s Model Mapping as well.
  • For lower-cost coding in Claude Code, deepseek-v4-pro works in place of glm-5.2, with deepseek-v4-flash on the Haiku row.
  • Claude models return at most 32,768 output tokens per response. Other Claude Code notes are in the Claude Code guide.
  • Prices are in the model catalog.

​
Other tools CC Switch manages

CC Switch also manages Claude Desktop and Grok Build, which use one provider at a time like the three tools above. OpenCode, OpenClaw, Hermes, Pi, MiniMax Code, Kimi Code, and Oh My Pi can keep several providers side by side. For these tools, add LaoZhang API with Custom Configuration under that tool. A universal provider syncs only to Claude Code, Codex, and Gemini CLI. Use a Claude Code-group key for providers that send Anthropic Messages requests and a default-group key for OpenAI-compatible ones.

​
FAQ

​
CC Switch shows “Environment Variable Conflicts Detected”

CC Switch shows this banner when your shell exports variables whose names start with ANTHROPIC, OPENAI, GEMINI, or GOOGLE_GEMINI. Whether they cause a problem depends on the tool. Claude Code: values CC Switch writes to settings.json take precedence over shell variables with the same name, so those are usually harmless. These still send requests elsewhere:
  • CLAUDE_CODE_USE_BEDROCK, CLAUDE_CODE_USE_VERTEX, or CLAUDE_CODE_USE_FOUNDRY set in your shell. Unset them and remove them from your shell profile.
  • A base URL or token in the project’s .claude/settings.json or .claude/settings.local.json. Project settings take precedence, so remove those keys.
Gemini CLI: when it loads ~/.gemini/.env, Gemini CLI keeps variables that already exist in the shell. An exported GOOGLE_GEMINI_BASE_URL or GEMINI_API_KEY therefore overrides what CC Switch writes. If you followed the Gemini CLI guide, your shell profile exports both. To fix it, remove those lines from ~/.zshrc, ~/.bashrc, or your Windows environment variables, then open a new terminal.

​
Claude Code returns a 503 “no available channels”

Check the Haiku model first. If it’s still claude-haiku-4-5-20251001 or empty, Claude Code’s background tasks request a model LaoZhang API doesn’t offer. Set it to a model from Choose models. If the error persists, check that Claude Code’s token is in the Claude Code group; see model availability.

​
Codex config differs from the Codex CLI guide. Does it matter?

No, both work.
  • The Codex CLI guide adds a provider named laozhang and reads the key from an environment variable through env_key.
  • CC Switch sets model_provider = "custom" and stores the key in config.toml itself.
Switching to the default provider in CC Switch restores the config you had before.

​
Connectivity check passes, but requests return a 401

Connectivity check doesn’t validate the key. Copy the full key again from the console, save the provider, and switch to it again. Also check for conflicting shell variables as described above. See Invalid API key or 404.

​
I changed the model with /model, but CC Switch still shows the old one

Changes made inside Claude Code, Codex, or Gemini CLI with /model aren’t written back to CC Switch. To change the default model for good, edit the provider in CC Switch and switch to it again.