Codex CLI

Why configuration cannot fix it

Current Codex CLI versions require a custom provider, meaning a named API connection, with wire_api = "responses". The former Chat Completions wire option is no longer supported by current Codex.

The shipped provod.ai public contract includes POST /v1/chat/completions, POST /v1/messages, model discovery, and documented image endpoints. It does not include POST /v1/responses. Pointing Codex at https://api.provod.ai/v1 therefore makes it call a missing endpoint; changing a model name or key cannot convert that protocol.

This status comes from the current generated API surface and the request format required by Codex, not from whether a model appears in GET /v1/models. A catalog entry proves model discovery for published endpoints; it does not add Responses support.

What evidence would change the status

This page can publish setup instructions only after a public Responses contract exists and a real Codex session verifies its required behavior, including response events, tool calls and results, cancellation, retry boundaries, public errors, and usage accounting. No release date is promised here.

Do not use old instructions that set wire_api = "chat", invent a /v1/responses proxy, or place a platform key into an unverified configuration.

Use a supported alternative now

If you only need to verify the API key and model, follow Chat Completions. A public error code is the client-facing error.code in an API response; preserve it when diagnosing a supported endpoint, but an endpoint-not-found result from Codex does not indicate model failure.

Troubleshooting