Quickstart: Codex

Codex reads extra model providers from its config file. Add this one and set the key in your shell.

~/.codex/config.toml

model = "glm-5.3-flash-uncensored"
model_provider = "uncensored-opencode"

[model_providers.uncensored-opencode]
name = "Uncensored Opencode"
base_url = "https://uncensoredopencode.com/v1"
env_key = "UNCENSORED_OPENCODE_API_KEY"
wire_api = "chat"

What to know

env_key names the variable Codex reads the key from, so the key is never in the file.
wire_api = "chat" uses chat completions. The API also serves /v1/responses with function tools; see Formats for what that route maps.
To change model, change the model line to any id on Models.

Run it

shell

export UNCENSORED_OPENCODE_API_KEY="sk-nc-..."
codex "Run the tests and fix the first failure"

If it does not start

A 401 on the first request means Codex found no key. The variable has to be set in the same shell that starts Codex, under the exact name in env_key.
A 400 that lists model ids means the model line has a typo. The message names every id the API accepts.
A 402 means the account, or the key's cap, cannot cover the request. The message says which and by how much.
If Codex sits waiting, the model is reasoning. The stream stays open while it does.