Codex CLI

Use Cortecs models with Codex CLI

Codex CLI is OpenAI's terminal-based coding agent. Connect it to Cortecs through the OpenAI-compatible Responses API.

Before you begin, create a Cortecs API key by following the Member quickstart and choose a tool-calling model from the model catalog.

1. Install Codex CLI

On macOS, Linux, or WSL:

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

Alternatively, install Codex CLI with npm:

npm install -g @openai/codex

See the Codex CLI documentation for othe platforms.

2. Add Cortecs

Create or update ~/.codex/config.toml:

model = "<MODEL_ID>"
model_provider = "cortecs"

[model_providers.cortecs]
name = "Cortecs"
base_url = "https://api.cortecs.ai/v1"
experimental_bearer_token = "<CORTECS_API_KEY>"
wire_api = "responses"

Replace <MODEL_ID> with the exact ID from the model catalog and <CORTECS_API_KEY> with your generated key. For more options, see the Codex configuration reference.

3. Start Codex CLI

Launch Codex within your desired project directory:

codex # To launch with the model set in the config
codex -m <MODEL_ID> # To launch with a different model

Run /status inside Codex to confirm that the active model and provider are set to Cortecs.

Troubleshooting

Web search rejection

Some of the provided models may reject Codex's web search tool. Disable web search for a single run when using an affected model:

codex --model "<MODEL_ID>" --config 'web_search="disabled"'

To disable it for every Codex session, add this top-level setting to ~/.codex/config.toml:

web_search = "disabled"

Place the setting outside [model_providers.cortecs], then restart Codex.

Unsupported reasoning effort

Reasoning levels vary by model and provider. For some models is necessary to provide the reasoning level upon launch.

Use a supported value for a single run:

codex --model "<MODEL_ID>" --config 'model_reasoning_effort="high"'

To set the value for every Codex session, add it as a top-level setting in ~/.codex/config.toml.