Codex CLI & Extension Setup (OpenAI API)

Overview

OpenAI's Codex CLI and the Codex VS Code extension talk to a model provider over the Responses API (/v1/responses). LLMsRelay serves that endpoint, so you can drive the OpenAI API through your existing LLMsRelay credits and a Codex-group key.

Step 1 — Create a Codex key

In the dashboard → API keys, create ansk-cs4-… key and select theCodex group. It looks likesk-cs4-…. Copy it once — it's shown only at creation.

Step 2 — Switch the tier to Codex

Confirm that the key group is Codex. This routes the key to the OpenAI API. Create a separate Basic or Pro key for Claude if both providers must stay active; the account balance remains shared.

Step 3 — Configure Codex CLI

Codex CLI reads ~/.codex/config.toml. Add LLMsRelay Codex as a model provider that uses the Responses API:

model = "gpt-5.5"
model_provider = "llmsrelay"

[model_providers.llmsrelay]
name = "LLMsRelay Codex"
base_url = "https://api.llmsrelay.com/v1"
env_key = "LLMSRELAY_API_KEY"
wire_api = "responses"

Export your Codex-group key as the provider env var, then launch Codex:

export LLMSRELAY_API_KEY="sk-cs4-your-key"
codex

Step 4 — Codex VS Code extension

The Codex VS Code extension shares the same ~/.codex/config.toml. Once the provider above is set and LLMSRELAY_API_KEY is exported in your shell environment, the extension picks up LLMsRelay Codex automatically. Select the gpt-5.5 model from the extension's model picker.

Step 5 — Test

You can verify the endpoint directly without the CLI:

curl https://api.llmsrelay.com/v1/responses \
  -H "Authorization: Bearer sk-cs4-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.5",
    "input": "Say hello in 3 words."
  }'

A successful response confirms the key, tier, and endpoint are wired correctly. If you get a model error, run GET /v1/models to confirm the model id.

Troubleshooting

• Unknown model — the key is not in the Codex group, or the model id is wrong. Check the group and GET /v1/models.
• 401 / unauthorized — the env var isn't exported in the shell that launched Codex.
• Endpoint not found — make sure the base URL ends in /v1 and the provider uses wire_api = "responses".