Skip to main content

Codex

Point Codex at the gateway Responses API. This page is a configuration sketch. A live Codex binary has not been run against the gateway (not verified in the compatibility matrix).

Config​

~/.codex/config.toml:

[model_providers.yabanin]
name = "Yabanin"
base_url = "https://gateway.example.internal/v1"
env_key = "YABANIN_API_KEY"
wire_api = "responses"

Put the key in the environment, not in the file:

export YABANIN_API_KEY='yb_…'

Endpoint​

POST /v1/responses. When the resolved provider is openai and the org has no deny tool policy, the gateway forwards the body unchanged. Otherwise it translates the call into chat completions.

Bring your own provider key (BYOK)​

Your org admin adds the upstream provider key once in the admin console (Settings → Provider keys — see BYOK). When the model routes through the org credential, the client needs no changes: keep the yb_… gateway key above and the gateway authenticates upstream with the org key.

Limits​

  • previous_response_id is rejected with 400 in the translation path.
  • reasoning.effort is dropped in translation.
  • Fixture coverage is not a live Codex run.

Check​

curl -sS -o /dev/null -w '%{http_code}\n' \
  -H "Authorization: Bearer $YABANIN_API_KEY" \
  https://gateway.example.internal/v1/models

200 means the gateway accepted the key. It does not start Codex.