Codex Setup

Codex uses local configuration files for provider settings and authentication. To use FlowAPI, configure both:

config.toml
auth.json

Restart Codex after editing both files.

Codex versions differ in their supported wire API. If your Codex build only uses the OpenAI Responses transport, the provider root should be https://api.flowapi.net. If it offers an OpenAI-compatible chat-completions transport, use the /v1 endpoint expected by that Codex version.

1. Edit config.toml

Find or create the Codex config file:

OS	Path
macOS / Linux	`~/.codex/config.toml`
Windows	`%userprofile%\.codex\config.toml`

Use this template:


model_provider = "FlowAPI"
model = "gpt-4o"
review_model = "gpt-4o"
model_reasoning_effort = "medium"
disable_response_storage = true
network_access = "enabled"
windows_wsl_setup_acknowledged = true
 
[model_providers.FlowAPI]
name = "FlowAPI"
base_url = "https://api.flowapi.net"
wire_api = "responses"
requires_openai_auth = true

Change model and review_model to a model ID available in your FlowAPI account.

2. Edit auth.json

Find or create the Codex auth file:

OS	Path
macOS / Linux	`~/.codex/auth.json`
Windows	`%userprofile%\.codex\auth.json`

Use this template:


{
  "OPENAI_API_KEY": "YOUR_FLOWAPI_API_KEY"
}

Replace YOUR_FLOWAPI_API_KEY with your real FlowAPI key.

3. Restart Codex

Quit Codex fully and open it again. Start a new session and confirm the selected model is the FlowAPI model you configured.

If Codex does not start

Check that config.toml is valid TOML.
Check that auth.json is valid JSON.
Confirm base_url matches the wire API used by your Codex version.
Confirm the configured model exists in FlowAPI.
Try a lower reasoning effort if the chosen model does not support medium, high, or xhigh.