Skip to main content

Codex

Codex can use a custom provider that speaks the Responses API. Configure Voxvey as a user-level provider and set the model to openai/gpt-5.5.

1. Set the token​

export VOXVEY_TOKEN="<access-token>"

2. Add a Codex provider​

Edit ~/.codex/config.toml.

model_provider = "voxvey"
model = "openai/gpt-5.5"

[model_providers.voxvey]
name = "Voxvey"
base_url = "https://api.voxvey.com/v1"
env_key = "VOXVEY_TOKEN"
wire_api = "responses"

Keep this in the user-level config. Codex ignores provider and base URL changes in project-local .codex/config.toml files.

3. Run Codex​

codex --model openai/gpt-5.5

For a one-off run:

codex exec --model openai/gpt-5.5 "Explain the current repository structure."

Built-in OpenAI provider shortcut​

If you only want to redirect Codex's built-in OpenAI provider to Voxvey, use openai_base_url instead of a custom provider.

openai_base_url = "https://api.voxvey.com/v1"
model = "openai/gpt-5.5"

The custom provider approach is easier to reason about when you switch between direct OpenAI and Voxvey frequently.

Verify​

Run:

codex --config model_provider='"voxvey"' --config model='"openai/gpt-5.5"'

Ask a short question. If authentication fails, confirm VOXVEY_TOKEN is present in the shell that starts Codex.