Codex CLI
Run OpenAI's Codex CLI on your Standard Compute budget. The launcher connects Codex for that session only. Your usual Codex configuration, sandbox and approval settings are preserved.
Current Codex releases only talk to custom providers through the OpenAI Responses API (POST /v1/responses); wire_api = "chat" and Chat Completions-only endpoints are no longer supported. Standard Compute serves /v1/responses at https://api.stdcmpt.com/v1.
1. Install Codex and the launcher
npm install -g @openai/codex
- macOS / Linux / WSL
- Windows (PowerShell)
curl -fsSL https://standardcompute.com/cli/install.sh | sh
irm https://standardcompute.com/cli/install.ps1 | iex
The launcher requires Node.js 20 or newer. Open a new terminal after installing so standardcompute is on your PATH.
2. Launch from your project folder
standardcompute launch codex
Choose an API key and approve the connection in your browser. Codex opens automatically. Use the same command each time.
3. Check the connection
Ask Codex to reply "OK", then check Request Logs for the request and the model that answered.
Models
Codex starts with standardcompute, which uses your smart routing settings. To pin one model, pass its exact ID from the Models page:
standardcompute launch codex -- --model anthropic/claude-sonnet-5.5
Anything after -- is passed straight to Codex. For example, to resume the last session:
standardcompute launch codex -- resume --last
Codex's built-in web search is disabled on this connection. The launcher also rejects flags that would override the connection it sets up, such as --config, --profile and --oss.
Manual configuration
If you can't use the launcher, add Standard Compute as a provider in ~/.codex/config.toml:
model = "standardcompute"
model_provider = "standardcompute"
[model_providers.standardcompute]
name = "Standard Compute"
base_url = "https://api.stdcmpt.com/v1"
env_key = "STANDARD_COMPUTE_API_KEY"
wire_api = "responses"
Then export your key and start Codex:
export STANDARD_COMPUTE_API_KEY=YOUR_API_KEY
codex
Troubleshooting
- Rotated your key or switching accounts: run
standardcompute login. Close running agents first. - Unexpected request errors: re-run the installer to update the launcher, and update Codex CLI.
- 404 errors with a custom config: check that
base_urlends in/v1andwire_api = "responses". - Payment errors (402): your monthly budget is used. Check it on the Usage page; upgrade or turn on Pay as you go from Billing.
- Request limits (429): wait for the
retry-afterdelay before retrying.