DEV Community

APIVAI
APIVAI

Posted on • Originally published at apivai.com

How to Use Codex CLI with a Cheap GPT API (Save 90%)

#codexcli #gptapi #tutorial

Use Codex CLI without paying full price

OpenAI's Codex CLI is a fast way to edit code, run commands, and pair with a model directly from your terminal. The friction is cost: pointed at official endpoints, a busy day of agentic coding adds up quickly. The good news is that Codex CLI talks to an OpenAI-compatible endpoint, so you can route it through a cheaper gateway with a one-line change and keep the exact same workflow.

This guide shows how to configure Codex CLI to use APIVAI, which resells GPT and Claude models at a steep discount off official pricing.

What you need

  • Codex CLI installed and working.
  • An APIVAI API key.
  • The APIVAI base URL: https://api.apivai.com/v1.

Step 1 — point Codex at the APIVAI endpoint

Codex CLI reads an OpenAI-style base URL and API key from environment variables. Set them in the shell that launches Codex.

export OPENAI_BASE_URL="https://api.apivai.com/v1"
export OPENAI_API_KEY="YOUR_APIVAI_API_KEY"
Enter fullscreen mode Exit fullscreen mode

On Windows PowerShell:

$env:OPENAI_BASE_URL="https://api.apivai.com/v1"
$env:OPENAI_API_KEY="YOUR_APIVAI_API_KEY"
Enter fullscreen mode Exit fullscreen mode

That is the whole change. Codex now sends its requests to APIVAI instead of the default endpoint, and your existing prompts, approvals, and file edits behave the same.

Step 2 — pick a model

Before hardcoding a model name, list what your account can use and choose one from the response.

curl https://api.apivai.com/v1/models \
  -H "Authorization: Bearer $OPENAI_API_KEY"
Enter fullscreen mode Exit fullscreen mode

Set the model Codex should use to one of the returned IDs, for example a GPT-class model for general coding. APIVAI exposes the OpenAI Responses API that Codex relies on, so tool calls and streaming work as expected.

Step 3 — run a small task first

Start with a tiny, low-risk request to confirm routing before turning Codex loose on a real change.

codex "explain what this repo does in two sentences"
Enter fullscreen mode Exit fullscreen mode

If you get a normal response, the endpoint, key, and model are all wired correctly. From here, use Codex exactly as you would against the official API.

Why route Codex through APIVAI

  • Lower effective cost on GPT and Claude models, with no subscription.
  • One OpenAI-compatible key that also works with Cursor, Claude Code, and your own scripts.
  • Pay-as-you-go top-ups, including crypto and Alipay.
  • Model discovery via /v1/models so you avoid stale, hardcoded model names.

Troubleshooting

  • A 401 usually means the key is missing or not exported in the shell that launched Codex.
  • A 404 often means the base URL is missing /v1 or has it twice.
  • A model error means the configured model is not in the current /v1/models response — pick a returned ID.
  • If streaming looks stuck, confirm the client reads the response incrementally and retry a non-streaming request to isolate the issue.

Get started

Create an API key, export the two environment variables above, choose a model from /v1/models, and run a small Codex task to confirm everything works. Most existing OpenAI-compatible setups need only the base-URL and key change.

Top comments (0)