Codex

Codex CLI is OpenAI's official terminal coding agent. TokenByte's OpenAI-compatible endpoint is a drop-in replacement for the default, and the recommended setup is to pin it in ~/.codex/config.toml plus auth.json.

Prerequisites

Node.js 20+ (download)
On Windows, install Git; it's optional on macOS / Linux
A TokenByte API key from the console

Configuration

Install Codex CLI

sudo npm install -g @openai/codex

Run from CMD or PowerShell:

npm install -g @openai/codex

Verify the install

codex --version

Output like codex-cli 0.41.0 means it's ready.

Wire in TokenByte

In the .codex folder under your home directory, create config.toml:

model = "<paste exact model title from /models>"

model_provider = "custom"

model_reasoning_effort = "high"

disable_response_storage = true

[model_providers.custom]

name = "custom"

base_url = "https://api.tokenbyte.ai/v1"

wire_api = "responses"

The model value must match the full card title shown on the Models page or in the Console → Models catalog.

Then create auth.json in the same .codex directory:

{
  "OPENAI_API_KEY": "<your API key>"
}

Replace <your API key> with the full key from Console → API Keys (e.g. sk-tb-...).

On Windows the .codex directory may not exist — create it manually. Explorer hides dot-directories by default; either enable View → Show → Hidden items or just run mkdir .codex.

Launch and sanity-check

From any project directory:

codex

Once the welcome screen appears, try:

Hi — please explain this code.

A real reply means TokenByte is live.

VS Code extension (optional)

Only connect the VS Code extension after the CLI works end-to-end — it narrows the debug surface.

Search the marketplace for Codex; only install the entry with the OpenAI verified badge.
The extension reads ~/.codex/config.toml and auth.json — nothing to re-enter in VS Code.

If codex still fails in the terminal, installing the extension won't magically fix it — config path and env var issues usually surface there first.

Stuck? See the FAQ.