Codex

Codex is OpenAI's official command-line coding agent and only speaks the OpenAI Responses protocol. Connect it by adding a custom model provider to its config file.

• Protocol: OpenAI Responses (/v1/responses)
• Base URL: https://tokenfunc.com/v1 (with /v1)
• Models: the GPT family, for example gpt-5

Option 1: one-click setup script (recommended)

Run the command below in a terminal, paste your key and pick a default model when asked. The script:

• Checks the key against the model list endpoint (free of charge)
• Installs Codex the official way if it is missing
• Adds this site as a model provider in ~/.codex/config.toml, keeping your other settings and backing up the original file
• Saves the key as the TOKENFUNC_API_KEY environment variable: in your shell startup file on macOS / Linux, as a user environment variable on Windows

curl -fsSL https://tokenfunc.com/install/codex.sh | bash

# Run in PowerShell (not Command Prompt)
irm https://tokenfunc.com/install/codex.ps1 | iex

When it finishes, open a new terminal (the key variable only applies to new terminals), then jump to the "Start and verify" step below.

Option 2: one-click import with CC Switch

With CC Switch installed, click the button below, make sure Codex is selected in the dialog and pick a model.

Enable the imported provider in CC Switch, then skip to the verify step below.

Option 3: manual setup

1. Install Codex

   macOS / Linux

   curl -fsSL https://chatgpt.com/codex/install.sh | sh

   Windows

   powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"

   npm

   npm install -g @openai/codex

2. Write the config file

   Add the content below to the Codex config file. If the file already has content, note that the model and model_provider lines must come first, before any [...] section; otherwise they are read as fields of the section above them and ignored.

   macOS / Linux

   ~/.codex/config.toml

   model = "gpt-5"
   model_provider = "tokenfunc"
   
   [model_providers.tokenfunc]
   name = "tokenfunc"
   base_url = "https://tokenfunc.com/v1"
   env_key = "TOKENFUNC_API_KEY"
   wire_api = "responses"
   requires_openai_auth = false

   Windows

   %USERPROFILE%.codex\config.toml

   model = "gpt-5"
   model_provider = "tokenfunc"
   
   [model_providers.tokenfunc]
   name = "tokenfunc"
   base_url = "https://tokenfunc.com/v1"
   env_key = "TOKENFUNC_API_KEY"
   wire_api = "responses"
   requires_openai_auth = false

   • env_key tells Codex to read the key from the TOKENFUNC_API_KEY environment variable, which you set next.
   • You can rename the provider (tokenfunc), but openai, ollama and lmstudio are reserved.

3. Set the key as an environment variable

   Codex recommends keeping the key out of the config file and in an environment variable.

   macOS / Linux

   # Save it in your shell profile so new terminals get it too (bash users: ~/.bashrc instead of ~/.zshrc)
   echo 'export TOKENFUNC_API_KEY="sk-xxxxxxxxxxxxxxxx"' >> ~/.zshrc
   source ~/.zshrc

   Windows

   # Saved permanently; open a new terminal afterwards
   setx TOKENFUNC_API_KEY "sk-xxxxxxxxxxxxxxxx"

4. Start and verify

   Open a new terminal, run codex and type /status. Check that the provider is tokenfunc and the model is the one you set. Then send "Reply with exactly: ok"; a reply means you are connected.

   Screenshot pendingThe /status output in Codex showing the tokenfunc provider and the current modeltools/codex-status.png

Choosing and switching models

• The default model comes from model in the config file.
• To switch once, run codex -m gpt-5, or type /model in a session.
• Codex relies on the Responses endpoint and tool calling. Some non-OpenAI models may not fully support them; if something looks wrong, switch back to a GPT model first.

Troubleshooting

Other errors: see FAQ and troubleshooting.