Codex

Codex CLI can connect to 4096bytes through an OpenAI compatible Provider. You need to write the dashboard-generated config.toml and auth.json into the Codex config directory. Prefer copying the config from the dashboard Use key dialog. It shows the correct path for your system and inserts the current API Key automatically.

Prerequisites

You have completed Create an account.
You have completed Get an API Key.
Codex CLI is installed and codex --version works.
The Use key dialog is open for the API Key you want to use.

The API Key in auth.json is stored as a local plaintext credential. Do not commit ~/.codex/auth.json, screenshots, or terminal output to a repository or chat.

Install

If Codex CLI is not installed, check the latest instructions in the OpenAI Codex GitHub repository. macOS or Linux:

curl -fsSL https://chatgpt.com/codex/install.sh | sh

Windows PowerShell:

powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"

If you use npm for global CLI tools:

npm install -g @openai/codex

After installation, reopen the terminal and verify:

codex --version

Copy config from the dashboard

Open API Keys

Open the dashboard and go to API Keys.

Choose Codex CLI

Find the key you want to use for Codex, click Use key, then choose Codex CLI.

Choose your system

Select macOS / Linux or Windows. The dashboard shows the target config file paths.

Copy config

Copy the config.toml and auth.json snippets. Put config.toml at the beginning of the file so old config does not override it.

macOS or Linux

Write `config.toml`

Open or create ~/.codex/config.toml, then put this block at the beginning.

model_provider = "OpenAI"
model = "gpt-5.5"
review_model = "gpt-5.5"
model_reasoning_effort = "xhigh"
disable_response_storage = true
network_access = "enabled"
windows_wsl_setup_acknowledged = true

[model_providers.OpenAI]
name = "OpenAI"
base_url = "https://api.4096bytes.com"
wire_api = "responses"
requires_openai_auth = true

[features]
goals = true

Use the model and review_model values shown in the dashboard. If your dashboard shows a different model name, use that value.

Write `auth.json`

Open or create ~/.codex/auth.json:

{
  "OPENAI_API_KEY": "YOUR_4096BYTES_API_KEY"
}

Replace YOUR_4096BYTES_API_KEY with the API Key you created in the dashboard.

Start Codex

Save both files, reopen the terminal, and run:

codex

You can also start with a low-risk prompt:

codex "Describe the purpose of the current directory in one sentence. Do not modify files."

Windows

Write `config.toml`

Open or create:

%USERPROFILE%\.codex\config.toml

Put this block at the beginning.

model_provider = "OpenAI"
model = "gpt-5.5"
review_model = "gpt-5.5"
model_reasoning_effort = "xhigh"
disable_response_storage = true
network_access = "enabled"
windows_wsl_setup_acknowledged = true

[model_providers.OpenAI]
name = "OpenAI"
base_url = "https://api.4096bytes.com"
wire_api = "responses"
requires_openai_auth = true

[features]
goals = true

If you run Codex inside WSL, configure ~/.codex/config.toml inside WSL instead of the Windows user directory.

Write `auth.json`

Open or create:

%USERPROFILE%\.codex\auth.json

Write:

{
  "OPENAI_API_KEY": "YOUR_4096BYTES_API_KEY"
}

If you run Codex inside WSL, write ~/.codex/auth.json inside WSL.

Start Codex

Save both files, reopen PowerShell or Windows Terminal, and run:

codex

You can also test with:

codex "Describe the purpose of the current directory in one sentence. Do not modify files."

Field reference

Field	Meaning
`model_provider`	Selects the Provider Codex should use. Here it must match `[model_providers.OpenAI]`.
`model`	Default model name. Copy the value shown in the dashboard.
`review_model`	Model used by Codex Review. Usually the same as `model`.
`base_url`	4096bytes API address. Use `https://api.4096bytes.com`.
`wire_api`	Uses the Responses API route. Set it to `responses`.
`requires_openai_auth`	Makes Codex read `OPENAI_API_KEY` from `auth.json`.
`OPENAI_API_KEY`	The 4096bytes API Key created in the dashboard.

Common issues

Codex still uses the old model or address

Put the new config at the beginning of config.toml. Save, close the terminal, and run codex again.

Codex reports authentication failure

Check that auth.json is valid JSON. Confirm the API Key has no missing characters, newlines, or extra spaces.

Codex cannot find the model

model and review_model must match the dashboard model name exactly. Do not copy names from someone else’s screenshot.

Windows config does not take effect

Confirm where Codex is running. If it runs in WSL, configure WSL’s ~/.codex/, not %USERPROFILE%\.codex\.

The config already has other Providers

You can keep them, but make sure model_provider = "OpenAI" points to the 4096bytes [model_providers.OpenAI] block.

Quickstart

Clients

SDKs

Prerequisites

Install

Copy config from the dashboard

macOS or Linux

Write `config.toml`

Write `auth.json`

Start Codex

Windows

Write `config.toml`

Write `auth.json`

Start Codex

Field reference

Common issues

​Prerequisites

​Install

​Copy config from the dashboard

​macOS or Linux

​Write config.toml

​Write auth.json

​Start Codex

​Windows

​Write config.toml

​Write auth.json

​Start Codex

​Field reference

​Common issues

Prerequisites

Install

Copy config from the dashboard

macOS or Linux

Write `config.toml`

Write `auth.json`

Start Codex

Windows

Write `config.toml`

Write `auth.json`

Start Codex

Field reference

Common issues