Codex

Prepare API Access

Have the three items below ready before you configure anything:

API Key

The key string you generated when creating your API key.

Base URL

https://api.watt-cloud.com/v1

The trailing /v1 is required — do not drop it.

Model Name

Take the complete model identifier from the Model Marketplace, for example DeepSeek-V4-Pro.

Installing Codex

Official Codex channels distribute the installer — see Codex CLI.

Linux / macOS / WSL2

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

Node.js 22+ users can take the npm route instead:

npm install -g @openai/codex

Windows (PowerShell)

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

Connect Codex to Watt Cloud

Requests travel via a custom model provider that you declare in ~/.codex/config.toml, the user-level configuration file of Codex.

Important: the model_provider and model_providers keys are honored only in ~/.codex/config.toml. A project-local .codex/config.toml gets them ignored by Codex.

Previously logged in via ChatGPT? Run the logout first so the two auth methods don't clash:

codex logout

Edit the Configuration File

Open ~/.codex/config.toml — create it if it doesn't exist yet — and add:

model = "GLM-5.1"
model_provider = "watt"

[model_providers.watt]
name = "Watt Cloud"
base_url = "https://api.watt-cloud.com/v1"
env_key = "WATT_API_KEY"
wire_api = "responses"
requires_openai_auth = false

Field Reference

Field	Description
model	The model Codex runs by default, e.g. GLM-5.1, DeepSeek-V4-Pro or MiniMax-M2.7
model_provider	ID of the provider; it has to match the [model_providers.<id>] block below
name	Human-readable label shown for this provider
base_url	The API endpoint — the trailing /v1 is mandatory
env_key	Environment variable that holds your API key
wire_api	Which API protocol to speak; choose responses for third-party endpoints
requires_openai_auth	Leave as false for custom OpenAI-compatible APIs

Set the API Key

Whichever variable env_key names is where Codex looks for the key, so export it before launching Codex:

export WATT_API_KEY="your-api-key"

To make the setting survive new sessions, drop that export line into the matching shell profile, e.g. ~/.bashrc or ~/.zshrc.

Change the Model

To run a different model, adjust the model value in ~/.codex/config.toml:

model = "DeepSeek-V4-Pro"

Verify the Connection

Once WATT_API_KEY is exported, try:

codex exec -m GLM-5.1 --sandbox read-only "Reply with exactly one word: hello"

Everything is wired up correctly when the output's session header includes provider: watt and the model then answers with pong.

Start Using the Model

From the directory of your project, launch Codex's interactive CLI:

codex