Codex and Codex++

Codex uses the OpenAI Responses protocol. Set the provider Base URL to the root domain; Codex then requests /v1/responses or another Responses path.

Codex++ interface

Create a custom provider in Codex++:

Field	Value
Provider name	Codex Relay
Base URL	https://codexapi.asia
API type	Responses
API Key	Your Codex Relay key
Model	An exact ID returned by /v1/models

Save the provider, switch to it, and open a new task. If the app caches models, refresh its model list or restart the app.

config.toml example

model_provider = "codex_relay"
model = "MODEL_ID"
model_reasoning_effort = "high"

[model_providers.codex_relay]
name = "Codex Relay"
wire_api = "responses"
requires_openai_auth = true
base_url = "https://codexapi.asia"

Store the API key in Codex++ credential management or the Codex authentication store. Do not write it directly into config.toml.

Fast / 1.5x mode

The client, gateway, selected model, and upstream account must all support the requested service tier:

service_tier = "fast"

[features]
fast_mode = true

A setting does not guarantee activation

fast_mode = true asks for the Fast service tier. If the model, group, or upstream account does not support it, the request may run in the standard mode. Verify the recorded service tier and measured latency rather than relying on the config file alone.

Troubleshooting

401 Unauthorized

Set the API key again for the active provider. Do not use the console login password as an API key.

Empty model list or model 404

Call /v1/models with the same key and set model to an exact returned ID.

The URL becomes /v1/v1/responses

Use https://codexapi.asia as the provider base_url; do not append /v1.

Fast mode does not look faster

Compare with a lower reasoning effort first. Time to first token also includes queueing, account-pool capacity, upstream networking, and tool preparation.