Use ApiClo as a custom provider in Codex

Before you start

• Use a current Codex CLI or IDE extension build and create an ApiClo API key.
• Check that the chosen ApiClo model is available through the Responses endpoint and review its tariff.

Steps

1. Edit the user config

   Open the user-level ~/.codex/config.toml file. Provider and authentication settings belong in the user config, not a project-level .codex/config.toml.

2. Add provider settings

   Set model_provider=apiclaude, base_url=https://apiclo.com/v1, env_key=APICLAUDE_API_KEY and wire_api=responses. Choose an exact ApiClo model ID.

3. Set the key and test

   Set APICLAUDE_API_KEY in the launching environment, restart Codex if needed, then ask for a short reply in a test folder and inspect ApiClo Logs.

Examples

model = "claude-sonnet-5"
model_provider = "apiclaude"

[model_providers.apiclaude]
name = "ApiClo"
base_url = "https://apiclo.com/v1"
env_key = "APICLAUDE_API_KEY"
wire_api = "responses"

export APICLAUDE_API_KEY="YOUR_APICLO_KEY"
codex

$env:APICLAUDE_API_KEY="YOUR_APICLO_KEY"
codex

Check the result

Codex returns a short response and the request appears in ApiClo usage under the selected model. Check tool use separately before relying on an agent workflow.

Troubleshooting

• 401: confirm that APICLAUDE_API_KEY is available to the Codex process and contains an active ApiClo key.
• 404 or model error: check the exact Model ID and the /v1 Base URL; do not append /responses to base_url.
• If Codex rejects the provider, update it and keep wire_api=responses, the supported custom-provider protocol.

Compatibility and billing

ApiClo is independent of OpenAI. Custom-provider behavior can vary by Codex version and surface; an API key is not a ChatGPT subscription. ApiClo charges model usage at the selected-model tariff.

Sources

• Codex configuration reference