1. Install Codex CLI

Shell
npm install -g @openai/codex
codex --version

2. Configure ~/.codex/config.toml

TOML
model = "YOUR_RESPONSES_MODEL_ID"
model_provider = "ghc-api"
model_reasoning_effort = "high"

[model_providers.ghc-api]
name = "ghc-api"
base_url = "http://127.0.0.1:8313/v1"
env_key = "GHC_API_TOKEN"
wire_api = "responses"
Shell
export GHC_API_TOKEN=1
codex

3. Match model capabilities

Select a model whose supported_endpoints contains /responses. Only set model_reasoning_effort to a value listed under capabilities.supports.reasoning_effort.

Do not assume every GPT model supports every effort.

Live model metadata is the source of truth for the current account.

4. Verify

  1. Confirm Responses support in /v1/models/full/.
  2. Check that requests arrive at /v1/responses.
  3. For 401 errors check the environment token; for 400 errors check the model and effort.
  4. Disable reverse-proxy buffering for streaming.

Codex configuration reference ↗

Next step

Inspect the models and endpoints available to your account before copying a model ID into client configuration.

Inspect models and parameters →