Codex + your provider
Give a Codex launch setup its own model provider, or use a native profile you already have.
Get the provider details
Install Codex and get your provider’s base URL, API key, and model ID. The endpoint must support the OpenAI Responses API; Chat Completions compatibility is not enough.
The example adds Codex · work without changing other Codex entries or editing Codex’s global configuration.
Already have a Codex profile? Skip to Use an existing Codex profile.
1. Define the provider
Add this setup to facade.json (see Agent settings). Replace the example base URL, key, and model ID:
{
"clis": {
"Codex · work": {
"type": "codex",
"model": "YOUR_MODEL_ID",
"args": [
"-c",
"model_provider=work",
"-c",
"model_providers.work.name=Work",
"-c",
"model_providers.work.base_url=https://api.example.com/v1",
"-c",
"model_providers.work.wire_api=responses",
"-c",
"model_providers.work.env_key=WORK_API_KEY"
],
"env": {
"WORK_API_KEY": "YOUR_API_KEY"
}
}
}
}Use the same provider ID, work, in model_provider=work and every model_providers.work.… setting. The key’s variable name, WORK_API_KEY, must match env_key.
2. Check the provider settings
| Setting | What it does |
|---|---|
model_provider=work | Selects the provider defined by the work settings. |
model_providers.work.name | Gives that provider a display name. |
model_providers.work.base_url | Sets the API base URL. |
model_providers.work.wire_api=responses | Uses the Responses API format. |
model_providers.work.env_key=WORK_API_KEY | Tells Codex which environment variable contains the key. |
env.WORK_API_KEY | Supplies that key to the Codex process. |
Each -c sets one Codex option for agents started with this entry.
The example stores the key in facade.json. If the key already lives in an environment variable, set env.WORK_API_KEY to ${env:WORK_API_KEY}. The editor must have that variable before the agent starts.
3. Start Codex with this setup
Save facade.json, choose New Agent, and select Codex · work. Use /model to check the selected model, then send a short task for your project.
If your provider reports usage, confirm that the request appears on the right account. To check the launch configuration, run Show Agent Launch and choose the working location. Find this entry in the report to see how the next agent would start with your current settings.
For request errors, see Troubleshooting.
Example: QwenCloud
For a QwenCloud Token Plan key, use https://token-plan.maas.qwencloudapi.com/compatible-mode/v1 as the base URL and keep wire_api=responses. Use a model available on your plan. Pay-as-you-go keys use the endpoint for their region.
Use an existing Codex profile
If you already use a native Codex profile, select it through the entry’s arguments:
{
"clis": {
"Codex · profile": {
"type": "codex",
"args": [
"--profile",
"work"
]
}
}
}Replace work with the profile you use. Keep the provider configuration in that profile instead of repeating it in the Facade entry. Save, choose New Agent, and select Codex · profile. Send a short task and confirm the request in your provider’s usage records, where available.