Codex CLI · custom provider

Codex CLI custom provider for GPT, Claude and Gemini

The Codex CLI talks to custom providers over the Responses API, which LLM Router supports. One installer command writes the provider and a profile per model family, so you switch models with a flag.

Create an account → Top up from $3

wire_api = "responses"supported

gpt · sonnet · geminigenerated profiles

15models in the catalog

Install

$ curl -fsSL https://llm-router.org/i/codex.sh | bash

PS> irm https://llm-router.org/i/codex.ps1 | iex

The installer asks for your API key, installs the Codex CLI if needed and writes ~/.codex/config.toml. It needs Node.js LTS.

Setup

Create an account

Email and password; no card is needed to register.

Create an account →

Top up your balance

From $3. You are billed only for the tokens you use.

Top up from $3 →

Create a key and run the installer

Create the key in the dashboard (it starts with llmr_), then run the installer and paste it. An existing config.toml is backed up before it is replaced.

macOS / Linux

curl -fsSL https://llm-router.org/i/codex.sh | bash

Windows

irm https://llm-router.org/i/codex.ps1 | iex

Start Codex with a profile

One profile per model family.

codex --profile gpt
codex --profile sonnet
codex --profile gemini

Payment

The balance is kept in US dollars and charged per token. Minimum top-up: $3. For payments in rubles the rate is shown on the top-up page before you pay.

SBP

QR code · credited in seconds

RU cards

Visa · Mastercard · MIR

International cards

Visa · Mastercard · issued outside RU

Cryptocurrency

USDT and others

Top up from $3 →

What the installer writes

Codex is pointed at LLM Router through the Responses API. Profiles are generated from the current catalog, so yours may differ from this example.

~/.codex/config.toml

model_reasoning_effort = "high"
model_provider = "llmrouter"

[model_providers.llmrouter]
name = "https://llm-router.org"
base_url = "https://llm-router.org/v1"
wire_api = "responses"

[profiles.gpt]
model = "gpt-…"

Worth knowing

Reasoning effort

model_reasoning_effort = "high" raises the thinking budget on models that support it. Use "medium", or remove the line, to spend less.

Re-running the installer

It saves the current config.toml as config.toml.bak.<date> first.

Prices

Per 1M tokens, USD. The full list is on the pricing page.

OpenAI GPT

8 models

gpt-6-astra $4.00 in · $20.00 out

gpt-5.6-sol $1.60 in · $8.00 out

gpt-5.6-terra $0.80 in · $4.80 out

Google Gemini

3 models

gemini-3.5-flash $0.60 in · $3.60 out

gemini-3.8-flash $0.33 in · $1.65 out

gemini-3-flash $0.20 in · $1.20 out

All models and prices →

Troubleshooting

command not found: codex

Node.js LTS is missing. Install it and run the installer again.

Profile not found

Profiles exist only for models in the current catalog. Check the [profiles.*] sections in config.toml.

401

The key is wrong or disabled. Copy it again from the dashboard.

402

Insufficient balance. Top up in the dashboard and retry.

Questions

How do I set up a custom provider in the Codex CLI?

Add a [model_providers.<id>] table with base_url and wire_api = "responses", and select it with model_provider. The installer writes this for LLM Router, together with your key.

Can I use Claude and Gemini models from Codex?

Yes. The installer adds profiles for them (gpt, sonnet, gemini), so you run codex --profile sonnet or --profile gemini.

How do I pay for usage?

Top up the balance by SBP, a Russian card (Visa, Mastercard, MIR), an international card or crypto. The minimum is $3, and usage is charged per token.

What happens when the balance runs out?

Requests are rejected with a 402 error until you top up. There is no subscription and no automatic charge.

Do you store my prompts and code?

No. We log metadata only: model, tokens and latency. Prompt content is not stored or used for training.

Create an account, top up, run one command

Registration needs an email and a password. The key is created in the dashboard once you are in.