Codex CLI

Codex CLI speaks the Responses API. Its configuration reference lists responses as the only wire protocol. PromptsForLess serves Responses at the same base URL as Chat Completions, so you only need a provider profile.

Before you startCopy Anchor Link

Install Codex CLI with its official setup instructions. Create an API key in the dashboard and store it as PFL_API_KEY.

Create a profileCopy Anchor Link

Merge this into your user-level ~/.codex/config.toml, keeping your existing settings:

[model_providers.promptsforless]
name = "PromptsForLess"
base_url = "https://api.promptsforless.com/v1"
env_key = "PFL_API_KEY"
wire_api = "responses"
requires_openai_auth = false
supports_websockets = false

[profiles.promptsforless]
model_provider = "promptsforless"
model = "gpt-6-luna"

A profile keeps PromptsForLess separate from your other providers, so you choose it when you want it. Codex ignores provider definitions in a project-level .codex/config.toml, so use your user config.

With PFL_API_KEY set in the same terminal, start Codex:

codex --profile promptsforless

Check the connectionCopy Anchor Link

Send a direct request to confirm your key and model:

curl --fail-with-body https://api.promptsforless.com/v1/responses \
  -H "Authorization: Bearer $PFL_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"model":"gpt-6-luna","input":"Reply with hello."}'

A completed response includes output text. Then start Codex in a small repository and ask it to describe a file.

Keep the protocol settingsCopy Anchor Link

Keep wire_api = "responses", because current Codex does not support chat. Keep supports_websockets = false so Codex streams over standard HTTP. OpenAI platform features such as hosted web search and file search stay on OpenAI's platform; changing the base URL does not bring them along.

TroubleshootingCopy Anchor Link

See the quickstart for direct HTTP checks and models for prices.