Codex

Codex uses the OpenAI Responses API. Configure the address and key, then select a model supporting Responses, streaming and tool calling.

1. Install and check the version

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

See the official Codex documentation  for system requirements and other installation methods.

2. Configure the Fenno provider

Back up and edit the user-level ~/.codex/config.toml. Merge these fields without replacing your MCP, permission or other provider settings.

model = "<YOUR_MODEL_ID>"
model_provider = "fenno"
 
[model_providers.fenno]
name = "Fenno"
base_url = "https://api.fenno.ai/v1"
env_key = "FENNO_API_KEY"
wire_api = "responses"
requires_openai_auth = false

Replace <YOUR_MODEL_ID> with an exact model ID that supports Responses. fenno is a local provider identifier; do not prepend it to the model value.

3. Set the key

Use Bash/Zsh on macOS, Linux or WSL, or the PowerShell example on Windows. The variable must be available in the same terminal that starts Codex:

export FENNO_API_KEY="<YOUR_FENNO_API_KEY>"

$env:FENNO_API_KEY = "<YOUR_FENNO_API_KEY>"

This configuration uses env_key. There is no need to place the Fenno key in auth.json, and it is not an OpenAI sign-in credential.

4. Verify a request

Start in an existing test Git repository with a request that does not modify files:

cd your-test-project
codex exec "Reply with OK only. Do not read or modify files."

Check the startup provider and model, then match the request time in the Fenno console. Test tool calling separately using a disposable file; a text response is not proof of full coding compatibility.

Troubleshooting

See OpenAI’s advanced configuration  and the general troubleshooting guide.