Codex CLI

Codex CLI is OpenAI’s official terminal coding assistant. One config.toml routes it through 爱玩Ai.

💡Recommended: skip the manual setup with CC Switch one-click import — it writes the Codex config for you and lets you switch between sites in one click.

Prerequisites

Node.js 18+ installed

An API key created on this site

Step 1: Install Codex CLI

npm install -g @openai/codex

Step 2: Write the configuration

Edit (or create)~/.codex/config.toml:

Windows: press Win + R, enter%userprofile%.codex

macOS / Linux: mkdir -p ~/.codex

#:schema https://developers.openai.com/codex/config-schema.json
model_provider = "custom"
model = "gpt-5.6"
model_reasoning_effort = "high"

[model_providers.custom]
name = "爱玩Ai"
base_url = "https://api.aiwanai.cc/v1"
wire_api = "responses"
requires_openai_auth = true

Unlike Claude Code, Codex’s base_url must include the /v1 suffix, and wire_api uses”responses”(the current and only supported value). The first schema line lets compatible editors flag invalid TOML keys.

gpt-5.6 is listed in both OpenAI’s official Codex documentation and this site’s current model list. This site may also offer suffixed routing models; those are site-specific names, not official OpenAI aliases. Before using one, confirm that it appears on the Pricing page and is enabled for your key group.

Then store the key in~/.codex/auth.json:

{
  "OPENAI_API_KEY": "sk-your-api-key"
}

Step 3: Verify

Restart the terminal and run inside any project:

codex

A working conversation means you’re connected; calls appear in this site’s Usage Logs.

Optional: enable web search

Recent Codex versions gate web search behind the top-level web_search key in config.toml (not inside [model_providers.custom]):

web_search = "live"

When enabled, Codex performs searches through this site’s POST /v1/alpha/search endpoint. Notes:

Searches are served by the site channels that support this endpoint; if your key’s group has none yet, searches fail while normal chat keeps working.

Searches are billed per call (the web_search tool price, $0.01/call by default) and appear as separate entries in Usage Logs; failed searches are not charged.

Stability: when one channel fails a search (overload, timeout, …), the site automatically retries on another channel — no action needed. A real live search can take tens of seconds; that is normal.

Remove the line, or set web_search = “disabled”, when you don’t need it.

If you manage Codex through CC Switch, the ccswitch:// import protocol cannot carry this key: this site’s import dialog offers a copy button for the line — paste it at the top level of the imported provider config. To apply it globally, put it into CC Switch’s Codex common-config snippet and enable “Apply Common Config” on each provider.

Troubleshooting

401 / Unauthorized: is the key inauth. JSON complete? (It does not belong in config.toml.)

404 / model not found: does themodelname exist on the Pricing page and in your key’s group?

Stream errors / timeouts: system proxies often interfere — try disabling them for the terminal.

Trusted sources and self-checks

This site’s live model list and groups: source of truth for site model names and key groups.

Codex config basics: official~/.codex/config. TOML location and baseline configuration.

Codex advanced configuration: custom-provider, base_url, and wire_api examples.

Codex complete configuration reference: every currently supported TOML key and value. Check this and the schema after upgrading Codex.