Codex, every model.
One provider block.
Codex CLI speaks the Responses dialect. LayerX1 serves it natively. One block in config.toml — or npx layerx1 setup --tool codex — and the same key that runs Claude Code now runs Codex on GLM, MiniMax, Kimi, DeepSeek or GPT.
A provider in config.toml.
base_url includes /v1. wire_api is responses. Empty replies on a reasoning model usually mean max_output_tokens is too low — thinking spends from it first.
npx layerx1 setup --tool codexmodel = "lx1-deepseek-v4-flash" model_provider = "layerx1" [model_providers.layerx1] name = "Layer X1" base_url = "https://api.layerx1.com/v1" experimental_bearer_token = "lx1_your_key" requires_openai_auth = false wire_api = "responses"
- 01
Get a key
Sign up at layerx1.com and copy an lx1_ key from the dashboard.
- 02
Add the LayerX1 provider
Run npx layerx1 setup --tool codex, or add a provider block in ~/.codex/config.toml with base_url https://api.layerx1.com/v1 and your key.
- 03
Pick a model
Set model to a catalog id such as lx1-deepseek-v4-flash or lx1-gpt-oss-120b.
- 04
Run Codex
Start the CLI. Raise max_output_tokens if a reasoning model returns an empty reply — thinking spends from that budget first.
- Can Codex CLI use models other than GPT?
- Yes. Codex talks the Responses dialect. LayerX1 serves it natively, so the same CLI can run GLM, MiniMax, Kimi, DeepSeek, Qwen or GPT behind one key.
- Do I need an OpenAI ChatGPT subscription?
- No. A LayerX1 key replaces the OpenAI backend. You keep the Codex CLI; you change the provider block.
- Is this a Codex coding plan?
- It is a coding plan that Codex can use. Same idea as OpenCode Go or a lab coding plan — a monthly pool and a key — except the key also drives Claude Code, Cline, OpenCode and the rest.
Claude Code in the morning. Codex at night.
The catalog does not change when the harness does. One lx1_ key, both dialects, the coding plan you already paid for.
Your agent stays the same.
Its inference gets better.
Start free · no card · Starter from $5/mo