Codex Responses provider

A dedicated provider entry for a company-controlled CLI route.

Codex uses the Responses interface. Define a named provider rather than relying on the default OpenAI account path. Install the official CLI for the workstation and verify its version before applying client-specific options.

Edit ~/.codex/config.toml, or %USERPROFILE%\.codex\config.toml on Windows:

model = "gpt-5.4"
model_provider = "b2b_company"

[model_providers.b2b_company]
name = "Company inference"
base_url = "https://api.b2b-llm.com/v1"
wire_api = "responses"
env_key = "B2B_API_KEY"

Set the secret in the launching environment:

export B2B_API_KEY='mg_live_...'
codex

On PowerShell, use $env:B2B_API_KEY = 'mg_live_...' for the current process. A value saved at user scope is available only to newly started processes. Keep credentials out of repositories and shell transcripts.

The provider key in model_provider must match the section name. Do not enable requires_openai_auth for this custom environment-key recipe. Optional reasoning controls should match the selected model's supported values. If the client reports an unknown model, refresh the account catalog rather than replacing the endpoint with a different project.