Codex CLI

Settings for using Echo Token Station models from OpenAI's Codex CLI. Codex CLI connects through the Responses API.

Codex CLI is OpenAI's AI coding tool. It runs in the terminal. Codex CLI connects to Echo using only the Responses API (/v1/responses).

Tested: Codex CLI 0.157.1 (2026-09-27, Echo's test environment)

Before you start

Have your Echo API key and the ID of the model you want to use ready. You can find model IDs in the model list
Codex CLI is a tool for OpenAI models. We have not tested how it behaves with models other than OpenAI's

Set up

Put your Echo API key in the environment variable ECHO_API_KEY

bash

export ECHO_API_KEY="sk-..."

Write the following in the Codex CLI settings file ~/.codex/config.toml. If you set the environment variable CODEX_HOME, use config.toml in that folder

toml

model = "gpt-6-sol"
model_provider = "echo"

[model_providers.echo]
name = "Echo Token Station"
base_url = "https://token.echohub.io/v1"
env_key = "ECHO_API_KEY"
wire_api = "responses"

Item	Meaning
`model`	The ID of the model to use
`model_provider`	The name of the provider settings to use. Make it the same as `echo` in `[model_providers.echo]` below
`name`	The name shown on screen
`base_url`	The base URL. Include `/v1`
`env_key`	The name of the environment variable to read the API key from
`wire_api`	The protocol. Set it to `"responses"`

If you set wire_api to "chat", Codex CLI 0.157.1 stops at startup with the error wire_api = "chat" is no longer supported.

Check that it works

In a terminal where ECHO_API_KEY is set, go to your working folder and run the following command. To try it in a folder that is not a Git repository, add --skip-git-repo-check

bash

codex exec "Reply with the single word hi."

If an answer is shown, you are done. The call also appears in 「呼び出し履歴」 (Call history) in the console

If you specify a model ID that Codex CLI does not know, the warning Model metadata for ... not found appears. The call continues anyway.

Troubleshooting

Symptom	Cause	Fix
`wire_api = "chat"` is no longer supported at startup	`wire_api` is `"chat"`	Set `wire_api = "responses"`
404 with an HTML page instead of JSON	`base_url` does not end with `/v1`	Change it to `https://token.echohub.io/v1`
401 and `Invalid API key`	The key is wrong, stopped or expired, or has used up its spending limit	Check the key's status in the console
404 and `Invalid URL (POST /v1/v1/responses)`	`base_url` has `/v1` twice	Change it to `https://token.echohub.io/v1`
403 and `Insufficient balance`	Insufficient balance	Top up
503 and `The model … does not exist or is not available`	The model ID is wrong, or Echo does not offer the model	Check the model ID in `model`

The list of error bodies is in Errors and fixes.

Codex CLI sends requests with store: false (telling the upstream API not to store the answer). Echo passes this value to the upstream unchanged.

Sources

What	Source	Checked
Connecting with the `config.toml` settings, the `wire_api = "chat"` error, the warning for unknown model IDs, sending `store: false`	Measured in Echo's test environment (Codex CLI 0.157.1), with the same relay settings as production and a server that returns test responses in place of the upstream API	2026-09-27
`model_provider` and the items of `[model_providers.<id>]` (`name`, `base_url`, `env_key`, `wire_api`), `CODEX_HOME`	https://developers.openai.com/codex/config-advanced	2026-09-27
`wire_api = "chat"` was removed	https://github.com/openai/codex/discussions/7782	2026-09-27